Case Study 2: Choosing the Wrong Chart Type — Lessons from Real Mistakes

The gallery in this chapter shows you what to use. This case study shows you what NOT to use — chart-type choices that appeared in real publications and that communicated badly because the chart type did not match the question. Each example is a learning opportunity. The mistakes are not stupid; they are natural consequences of choosing a chart type by habit rather than by analysis. The gallery exists so you can make these choices deliberately.

Mistake 1: Pie Chart with 15 Categories

A government report showed the budget breakdown across 15 departments as a pie chart. Most slices were so thin they were unreadable. The labels overlapped. The reader could not tell whether Education or Defense was larger without reading the numbers.

Why it happened: the default chart type for "parts of a whole" is often a pie chart, and the author did not think about whether 15 categories was too many for the format.

The fix: a horizontal bar chart sorted by budget size. Each department gets a readable bar. The reader can compare magnitudes precisely. The chart takes the same space but conveys far more information.

Gallery reference: Section 35.3 (Composition) — the pie chart entry specifically says "when NOT to use: more than 5-7 categories."

Mistake 2: Line Chart for Unordered Categories

A marketing dashboard showed "customer satisfaction by product category" as a line chart. The x-axis was "Electronics, Clothing, Food, Home, Sports" — categories with no natural order. The line implied a trend from left to right that did not exist.

Why it happened: the dashboard tool defaulted to a line chart for numeric y-values, and the author did not override the default.

The fix: a bar chart (horizontal, sorted by satisfaction score). Categories have no natural order, so a line (which implies continuity) is misleading. Bars treat each category independently.

Gallery reference: Section 35.2 (Comparison) — bar charts are the default for categorical comparisons. Line charts are in Section 35.6 (Trend) and should only be used when the x-axis has a meaningful order.

Mistake 3: 3D Bar Chart Hiding Real Values

A corporate presentation showed quarterly revenue as a 3D perspective bar chart. The 3D effect made the bars look different sizes depending on their position — bars in the "back" appeared smaller than bars in the "front" even when their values were similar. The reader could not make accurate comparisons.

Why it happened: the 3D effect was chosen for visual impact. The presentation author thought it looked more impressive than a flat chart. The 3D rendering did not add information; it distorted it.

The fix: a standard 2D bar chart. All bars are the same visual depth. The reader can compare heights directly without perspective distortion.

Gallery reference: none of the 50 gallery entries use 3D effects, because 3D adds no information to 2D data and always distorts perception. This is a recurring theme since Chapter 6 (data-ink ratio).

Mistake 4: Scatter Plot for Time Series

An analyst plotted "stock price vs. date" as a scatter plot with individual dots, no lines connecting them. The temporal ordering was invisible. The reader could not see the trend — just a cloud of dots that happened to have dates on the x-axis.

Why it happened: the analyst used plt.scatter instead of plt.plot, probably because the code was faster to write or because the data had some missing days.

The fix: a line chart (with optional markers for individual points). The line connects the temporal ordering and makes trends visible. Missing days can be handled with df.resample or by leaving gaps in the line.

Gallery reference: Section 35.6 (Trend) — line chart is the canonical choice for temporal data. Scatter is in Section 35.5 (Relationship) and should be used when the x-axis is not time.

Mistake 5: Dual-Axis Chart Manufacturing a Correlation

A news article showed "ice cream sales" and "drowning deaths" on the same chart with different y-axes. The two lines appeared to track each other perfectly, and the article implied a causal link. In reality, both variables are correlated with temperature (a confounding variable), and the dual-axis scales were chosen to make the correlation look visually dramatic.

Why it happened: the author wanted to show two variables together and chose dual-axis for convenience. The scale choices were made to maximize visual overlap, which is the fundamental problem with dual-axis charts.

The fix: (1) a scatter plot of ice cream sales vs. drowning deaths, which shows the actual bivariate relationship without scale manipulation. (2) Two separate charts, stacked vertically with a shared x-axis, which shows both series without implying a correlation. (3) A note about the confounding variable (temperature).

Gallery reference: dual-axis charts are not in the gallery because they are inherently prone to misuse (Chapter 4, Chapter 21). The gallery recommends scatter plots (Section 35.5) for relationships and separate-panel line charts (Section 35.6) for comparing trends.

Mistake 6: Choropleth Without Normalization

A social media post showed a US county map of "total COVID cases" colored from light to dark. The map looked dramatic — Los Angeles, New York, and Chicago were deep red. But the map was just showing where people live. Total cases without per-capita normalization is a population map, not a COVID map.

Why it happened: the author did not normalize because the raw numbers looked more dramatic. Or they did not know about the normalization issue.

The fix: normalize to per-capita rates (cases per 100,000 residents). The map then shows actual incidence rather than population.

Gallery reference: Section 35.7 (Geospatial) — the choropleth entry specifically warns about the population-proxy pitfall and recommends per-capita normalization.

The Pattern

All six mistakes share a common cause: the chart type was chosen by habit, convenience, or aesthetic preference rather than by analyzing the question and the data. The gallery exists to break this habit. When you look up a chart type in the gallery, the "when to use" and "when NOT to use" entries force you to think about whether the type fits. The code template then gives you a fast path to the correct chart.

The gallery is not a list of chart types. It is a decision tool. Use it as one.

Discussion Questions

1. On chart-type habits. Which chart type do you default to most often? Is that default always appropriate?
2. On the 3D effect. Why does 3D persist in presentations despite decades of criticism?
3. On dual-axis charts. The gallery excludes dual-axis charts. Is this the right call?
4. On normalization. The choropleth mistake is one of the most common. How would you teach normalization to a non-technical audience?
5. On self-awareness. Have you made any of these six mistakes in your own work?
6. On the gallery as prevention. Would consulting the gallery before choosing a chart type prevent most of these mistakes?

Wrong chart types are not always obvious to their authors. The six examples in this case study were published in real reports, articles, and dashboards by competent people. The mistakes were natural consequences of choosing by habit rather than by analysis. The gallery is your defense against these mistakes — consult it before building, not after publishing.