I2 · Working with Data: NumPy, pandas, SQL and Visualisation

Data Storytelling with Charts

Data storytelling is communicating insights from data so that an audience understands and acts on them. It combines three elements: the data (accurate, relevant evidence), visuals (charts designed to make the key point obvious) and narrative (a clear message, context and recommended action). Good data stories know their audience, state the insight in the chart title, remove clutter, use colour and annotation to direct attention, show comparisons and context, and end with what should happen next.

12 min Intermediate · Sublevel Beginner Jump to the activity

What you need: The earlier topics of this module.

Definition

Two analysts show the same data to a district collector. One shows a dense chart with twelve colours and says "here is the data". The other shows one clear chart titled "Dengue cases rise four weeks after the monsoon peaks" and recommends when to start fogging. Only one changes a decision.

Data storytelling is communicating insights from data so that an audience understands and acts on them. It combines three elements: the data (accurate, relevant evidence), visuals (charts designed to make the key point obvious) and narrative (a clear message, context and recommended action). Good data stories know their audience, state the insight in the chart title, remove clutter, use colour and annotation to direct attention, show comparisons and context, and end with what should happen next.

In simple words

A chart is not finished when it shows the data; it is finished when someone looking at it for five seconds knows what you want them to know. Data storytelling is designing for those five seconds.

Overview

  • Data storytelling combines data, visuals and narrative to help an audience understand and act.
  • Start from the audience and the one message they must remember.
  • Choose the chart that fits the message, then remove clutter and highlight what matters.
  • A clear headline title states the finding, not just the topic.
  • Honest charts start bar axes at zero, show uncertainty and avoid cherry-picked ranges.
  • Colour should be used sparingly and be readable for colour-blind viewers.

Why it matters

Analysis only matters if it influences decisions. Leaders, citizens and clients are busy and rarely read tables. Clear data stories get resources to where they are needed, and misleading ones can do real harm, so honesty matters as much as clarity.

Key terms

Key message
The single insight the audience should take away.
Action title
A chart title that states the insight, such as "Cases peak four weeks after rain".
Decluttering
Removing borders, heavy gridlines, extra colours and labels that do not help.
Pre-attentive attributes
Features such as colour, size and position that the eye notices instantly.
Annotation
Text or arrows on a chart that explain a key point or event.
Context
Comparisons such as last year, the target or the national average that give numbers meaning.

Types

Data stories are built around a few types of message. Each type suits particular charts.

1. Change over time

A story about how something rose, fell or turned.

  • How it works: Use a line chart with the time axis left to right, annotate key events and highlight the period that matters.
  • Example: "Delhi's PM2.5 doubles every November", with the festival and stubble-burning weeks marked.
  • When to use it: Use it for trends, seasonality and before-and-after comparisons.

2. Comparison and ranking

A story about which items are bigger, smaller, better or worse.

  • How it works: Use sorted bar charts starting at zero, with the item of interest in a strong colour and the rest in grey.
  • Example: "Our district ranks 4th of 12 in vaccination coverage".
  • When to use it: Use it to show where something stands among peers.

3. Part-to-whole

A story about how a total is made up.

  • How it works: Use a stacked bar or, for two or three parts, a simple pie or donut; a treemap works for many parts.
  • Example: "Two-thirds of the household budget goes on food and rent".
  • When to use it: Use it for shares and composition, keeping the number of parts small.

4. Relationship

A story about how two things move together.

  • How it works: Use a scatter plot, perhaps with a trend line, and label the few points the audience should look at.
  • Example: "States that spend more per student have higher reading scores", noting that correlation is not proof of cause.
  • When to use it: Use it to show associations and outliers; be careful not to imply causation.

5. Distribution and uncertainty

A story about spread, inequality or how sure we are.

  • How it works: Use histograms, box plots or ranges with error bars, and say in words what the spread means.
  • Example: "Most deliveries take 25–35 minutes, but one in ten takes over an hour".
  • When to use it: Use it when averages hide important variation or when forecasts are uncertain.

Explained in detail

From exploration to explanation

Exploratory charts help you find insights; explanatory charts help others understand them. During EDA you may make fifty charts; for a presentation you choose the two or three that matter and redesign them for the audience. Cole Nussbaumer Knaflic's book Storytelling with Data (2015) sets out a practical process: understand the context and audience, choose an effective visual, eliminate clutter, focus attention where you want it, think like a designer and tell a story with a beginning (the situation), middle (the insight and evidence) and end (the recommended action).

Design choices that guide attention

TechniqueExample
Action title"Dengue cases peak in September, four weeks after the heaviest rain"
Highlight colourThe district of interest in blue, others in grey
AnnotationAn arrow and note: "Fogging started here last year"
Direct labelsLabel lines at their ends instead of using a separate legend
Context lineA dashed line for the state average or target
DeclutterRemove top and right borders, lighten gridlines, round numbers

Honesty in data stories

A persuasive chart can mislead as easily as it informs. Do not cherry-pick a time window that happens to support your message; show enough history for context. Keep scales consistent when comparing charts side by side. Report uncertainty, such as a range or the size of the sample, especially for small groups. When percentages are involved, show the base number too: "cases doubled" may mean a rise from 2 to 4. Separate correlation from causation in your wording, using "rose at the same time as" rather than "caused" unless you have evidence. Credit your data sources on the chart. A trustworthy story includes the caveats that matter; leaving them out may win one meeting but loses credibility over time.

Structure matters beyond a single chart. In a report or presentation, put the conclusion first, then the evidence, rather than building slowly towards it; busy readers may stop after the first page. Each chart should support one point in that argument, appear in a sensible order and connect to the next with a sentence of explanation. Ending with a clear, specific recommendation, who should do what by when, turns an interesting analysis into a decision.

How it works

  1. Identify your audience and the decision they face.
  2. Write the key message in one sentence and make it the chart title.
  3. Choose the simplest chart that shows the evidence, and remove clutter.
  4. Use colour, annotation and context lines to focus attention.
  5. End with a recommended action, and state caveats and sources honestly.

In code

storytelling_chart.py
import matplotlib.pyplot as plt
import numpy as np

weeks = np.arange(1, 53)
rain = np.clip(120 * np.exp(-((weeks - 30) / 5) ** 2) + np.random.default_rng(2).normal(0, 2, 52), 0, None)
cases = np.clip(90 * np.exp(-((weeks - 34) / 4.5) ** 2) + 5 + np.random.default_rng(3).normal(0, 1.5, 52), 0, None)

fig, (before, after) = plt.subplots(1, 2, figsize=(13, 4))
# Before: everything shouted at once
before.plot(weeks, rain, label="rain"); before.plot(weeks, cases, label="cases")
before.grid(True); before.legend(); before.set_title("Data")

# After: one message, highlighted and annotated
peak_rain, peak_cases = weeks[rain.argmax()], weeks[cases.argmax()]
after.plot(weeks, cases, color="crimson", linewidth=2.5)
after.plot(weeks, rain / rain.max() * cases.max(), color="lightgrey", linewidth=1.5)
after.annotate("Heaviest rain", (peak_rain, cases.max() * 0.95), (peak_rain - 14, cases.max() * 0.8),
               arrowprops={"arrowstyle": "->", "color": "grey"}, color="grey")
after.annotate(f"Cases peak {peak_cases - peak_rain} weeks later", (peak_cases, cases.max()),
               (peak_cases + 3, cases.max() * 0.7), arrowprops={"arrowstyle": "->"}, color="crimson")
for side in ["top", "right"]:
    after.spines[side].set_visible(False)
after.set_title(f"Dengue cases peak {peak_cases - peak_rain} weeks after the heaviest rain")
after.set_xlabel("Week of the year"); after.set_ylabel("Weekly dengue cases (example data)")
fig.tight_layout()

print("Rain peaks in week", peak_rain, "and cases in week", peak_cases)
print(f"Recommendation: start prevention by week {peak_rain}, about {peak_cases - peak_rain} weeks before the case peak")

Output

Rain peaks in week 30 and cases in week 34
Recommendation: start prevention by week 30, about 4 weeks before the case peak

Try it yourself

Exploratory or explanatory?

Which purpose does each chart serve?

  • A slide comparing this year with the target

  • Fifty quick histograms of every column

  • A scatter matrix to look for relationships

  • One annotated chart with an action title for the collector

Getting fogging started earlier

A district health officer needs to convince the municipal council to start dengue prevention earlier.

Common mistakes

  • Titles that describe the chart instead of stating the insight.
  • Too many colours, so nothing stands out.
  • Cherry-picking time ranges or hiding uncertainty to make a point.

Check yourself

Question 1

1. What is an action title?

Question 2

2. Why grey out most lines and colour one?

Question 3

3. Which is an honest practice?

Summary

  • Data storytelling combines accurate data, focused visuals and a clear narrative ending in action.
  • Use action titles, decluttering, highlight colour, annotation and context.
  • Stay honest: full context, consistent scales, base numbers, uncertainty and careful causal language.

Finished reading?

Sources: Cole Nussbaumer Knaflic, Storytelling with Data · Claus O. Wilke, Fundamentals of Data Visualization