←Back to the bookChapter 14 · Charts for Categorical Data

Part III · Visualizing Data  |  Chapter 14

Charts for Categorical Data 🐍 Notebook

Five demos on showing how a total splits across categories: the bar chart and its variants, the truncated-axis trick that fools the eye, and why a sorted bar usually beats a pie.

Author: John Fisher  ·  Statistics, Data Science and AI: A Visual Handbook  ·  2026

🎯 What you'll build in this notebook

#	Demo	Idea it builds
1	The workhorse bar chart	counts per category from value_counts
2	Sorted, horizontal bars	the fix for many or long labels
3	Grouped vs stacked	two ways to add a second category
4	How a chart lies	a truncated y-axis exaggerates differences
5	Pie vs bar	why length beats angle for reading values

---

⚙️ Setup, imports & the book's plotting style

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt

rng = np.random.default_rng(14)

NAVY="#0a1230"; INK="#1a2138"; INK_SOFT="#4a5578"
CYAN="#0891b2"; PURPLE="#7c3aed"; AMBER="#d97706"; GREEN="#059669"; PINK="#db2777"; BLUE="#2563eb"; GRID="#e6e9f2"
# a colorblind-aware categorical palette (Okabe-Ito inspired)
CATS=[CYAN, AMBER, PURPLE, GREEN, PINK, BLUE]
plt.rcParams.update({
    "figure.facecolor":"white","axes.facecolor":"white","figure.dpi":110,"font.size":11,
    "axes.edgecolor":GRID,"axes.linewidth":1.2,"axes.grid":True,"grid.color":GRID,"axes.axisbelow":True,
    "axes.spines.top":False,"axes.spines.right":False,"axes.titlesize":13,"axes.titleweight":"bold","axes.titlecolor":INK,
    "axes.labelcolor":INK_SOFT,"axes.labelsize":11.5,"xtick.color":INK_SOFT,"ytick.color":INK_SOFT,"legend.frameon":False,
})
print("✅ Environment ready.")

✅ Environment ready.

DEMO 1 · THE WORKHORSE

📊 A bar chart from counts

Categorical data answers "how many in each category?" The bar chart encodes that count as a bar length on a common zero baseline, the encoding the eye reads most accurately. Note the gaps between bars: that signals distinct categories.

browsers = pd.Series(rng.choice(["Chrome","Safari","Edge","Firefox","Other"],
                     size=300, p=[0.55, 0.22, 0.12, 0.07, 0.04]), name="browser")
counts = browsers.value_counts()

fig, ax = plt.subplots(figsize=(8, 4.5))
ax.bar(counts.index, counts.values, color=CYAN, edgecolor="white", width=0.7)
ax.set_title("Browser share of 300 visits"); ax.set_ylabel("count")
ax.set_ylim(0, counts.max()*1.15)        # bars start at ZERO
for i,v in enumerate(counts.values): ax.text(i, v+2, str(v), ha="center", fontweight="bold", color=INK)
plt.tight_layout(); plt.show()
print(counts.to_string())

browser
Chrome     154
Safari      78
Edge        38
Firefox     19
Other       11

DEMO 2 · SORTED & HORIZONTAL

📏 The fix for long or many labels

When labels are long or there are many categories, swap the axes: horizontal bars read like a ranked list. For unordered (nominal) categories, sorting by value makes comparisons effortless.

depts = pd.Series({"Customer Support":142,"Engineering":98,"Sales":76,
                   "Marketing":54,"Human Resources":31,"Finance":27})
depts = depts.sort_values()                # ascending -> biggest ends on top in barh

fig, ax = plt.subplots(figsize=(8, 4.5))
ax.barh(depts.index, depts.values, color=PURPLE, edgecolor="white", height=0.7)
ax.set_title("Tickets opened by department"); ax.set_xlabel("tickets")
ax.set_xlim(0, depts.max()*1.12)
for i,v in enumerate(depts.values): ax.text(v+2, i, str(v), va="center", fontweight="bold", color=INK)
ax.grid(axis="y", visible=False)
plt.tight_layout(); plt.show()
print("Horizontal + sorted: no rotated labels, instant ranking.")

Horizontal + sorted: no rotated labels, instant ranking.

DEMO 3 · GROUPED VS STACKED

🧱 Adding a second category

To split each category by a second variable you have two choices. Grouped bars share one baseline, so series compare precisely. Stacked bars show the total per category, but only the bottom segment sits on the baseline, so the floating upper segments are hard to compare.

quarters = ["Q1","Q2","Q3","Q4"]
online = np.array([120, 135, 150, 180])
store  = np.array([90,  85,  80,  70])
x = np.arange(len(quarters)); w = 0.38

fig, (a1, a2) = plt.subplots(1, 2, figsize=(13, 4.4))

a1.bar(x-w/2, online, w, label="Online", color=CYAN, edgecolor="white")
a1.bar(x+w/2, store,  w, label="In-store", color=AMBER, edgecolor="white")
a1.set_title("Grouped: compare the two series"); a1.set_xticks(x); a1.set_xticklabels(quarters); a1.legend()

a2.bar(x, online, label="Online", color=CYAN, edgecolor="white")
a2.bar(x, store, bottom=online, label="In-store", color=AMBER, edgecolor="white")
a2.set_title("Stacked: read the total per quarter"); a2.set_xticks(x); a2.set_xticklabels(quarters); a2.legend()
for ax in (a1,a2): ax.set_ylabel("sales"); ax.set_ylim(0, 270)
plt.tight_layout(); plt.show()
print("Grouped answers \"which channel is bigger?\"; stacked answers \"what is the quarter total?\"")

Grouped answers "which channel is bigger?"; stacked answers "what is the quarter total?"

DEMO 4 · HOW A CHART LIES

🚨 The truncated y-axis

Bars encode length from zero, so the value axis must start at 0. Cut the baseline and a tiny gap becomes a chasm. Same two numbers, two very different stories.

labels = ["Plan A", "Plan B"]; vals = [35.0, 39.6]

fig, (a1, a2) = plt.subplots(1, 2, figsize=(11, 4.4))
a1.bar(labels, vals, color=PINK, edgecolor="white", width=0.6)
a1.set_ylim(34, 40)                       # TRUNCATED: misleading
a1.set_title("Truncated axis (misleading)"); a1.set_ylabel("rate (%)")

a2.bar(labels, vals, color=GREEN, edgecolor="white", width=0.6)
a2.set_ylim(0, 45)                        # honest zero baseline
a2.set_title("Zero baseline (honest)"); a2.set_ylabel("rate (%)")
for ax in (a1,a2):
    for i,v in enumerate(vals): ax.text(i, v+0.1 if ax is a1 else v+0.6, f"{v}%", ha="center", fontweight="bold", color=INK)
plt.tight_layout(); plt.show()

true_ratio = vals[1]/vals[0]
looks_ratio = (vals[1]-34)/(vals[0]-34)
print(f"Real difference: {vals[1]/vals[0]:.2f}x  ({vals[0]}% vs {vals[1]}%)")
print(f"The truncated chart makes Plan B look about {looks_ratio:.1f}x Plan A. That gap is manufactured by the axis.")

Real difference: 1.13x  (35.0% vs 39.6%)
The truncated chart makes Plan B look about 5.6x Plan A. That gap is manufactured by the axis.

DEMO 5 · PIE VS BAR

🥧 Length beats angle

A pie asks the eye to compare angles and areas, which we judge poorly. The same shares as a sorted bar chart, read off a common scale, are far easier to rank. If you must use a pie: few slices, sorted, summing to 100%, directly labeled.

share = pd.Series({"Product A":31,"Product B":27,"Product C":24,"Product D":18})

fig, (a1, a2) = plt.subplots(1, 2, figsize=(12, 4.6))
a1.pie(share.values, labels=share.index, autopct="%1.0f%%", startangle=90,
       colors=CATS[:len(share)], wedgeprops=dict(edgecolor="white", linewidth=1.5))
a1.set_title("Pie: which two are nearly tied?")

s = share.sort_values()
a2.barh(s.index, s.values, color=BLUE, edgecolor="white", height=0.65)
a2.set_title("Sorted bar: the ranking is obvious"); a2.set_xlabel("market share (%)")
a2.set_xlim(0, 35); a2.grid(axis="y", visible=False)
for i,v in enumerate(s.values): a2.text(v+0.5, i, f"{v}%", va="center", fontweight="bold", color=INK)
plt.tight_layout(); plt.show()
print("B (27) vs C (24) are hard to separate on the pie, trivial on the bar.")

B (27) vs C (24) are hard to separate on the pie, trivial on the bar.

👥 Real-World Example: An Employee Survey

A survey of 400 employees, all categorical. First a plain bar chart ranks the departments by headcount. Then a grouped bar of satisfaction by work mode does the real work: it lines up the Low/Medium/High counts side by side for each arrangement, and the pattern jumps out, remote and hybrid staff report High satisfaction far more often than onsite staff.

# --- Real-World beat: bar chart and grouped bar on survey categories ---
BASE = "https://raw.githubusercontent.com/johnfisher-ai/Statistics-Data-Science-AI-Visual-Book/main/data/"
try:    survey = pd.read_excel("../../data/charts-for-categorical-data--employee_survey.xlsx", sheet_name="Data")
except FileNotFoundError: survey = pd.read_excel(BASE+"charts-for-categorical-data--employee_survey.xlsx", sheet_name="Data")
dept = survey.department.value_counts()
ct = pd.crosstab(survey.work_mode, survey.satisfaction)[["Low","Medium","High"]]
print("department headcount:", dept.to_dict())
print("satisfaction by work mode:"); print(ct.to_string())
fig, ax = plt.subplots(1, 2, figsize=(11.5,4.2))
ax[0].barh(dept.index[::-1], dept.values[::-1], color=CYAN, alpha=0.85, edgecolor="white")
ax[0].set_title("Headcount by department (sorted)"); ax[0].set_xlabel("employees")
modes = ["Onsite","Hybrid","Remote"]; levels = ["Low","Medium","High"]; cols = [AMBER, PURPLE, CYAN]
x = np.arange(len(modes)); w = 0.26
for j, lvl in enumerate(levels):
    ax[1].bar(x + (j-1)*w, [ct.loc[m, lvl] for m in modes], w, label=lvl, color=cols[j], edgecolor="white")
ax[1].set_xticks(x); ax[1].set_xticklabels(modes); ax[1].legend(title="satisfaction")
ax[1].set_title("Satisfaction by work mode (grouped)"); ax[1].set_ylabel("employees")
plt.tight_layout(); plt.show()

department headcount: {'Engineering': 113, 'Sales': 96, 'Support': 85, 'Marketing': 61, 'Operations': 45}
satisfaction by work mode:
satisfaction  Low  Medium  High
work_mode                      
Hybrid         17      56    68
Onsite         47      48    27
Remote         19      51    67

---

🎓 Recap

• The bar chart is the default for categories; bars have gaps and must start at zero.
• Horizontal, sorted bars handle long labels and many categories cleanly.
• Grouped bars compare series precisely; stacked bars show totals but hide inner comparisons.
• A truncated axis exaggerates differences, one of the most common ways charts mislead.
• A sorted bar usually beats a pie: we read length far better than angle.

Statistics, Data Science and AI: A Visual Handbook · © 2026 John Fisher

←Back to the bookChapter 14 · Charts for Categorical Data