18. Misleading charts — conveying a falsehood without lying · Dr. Seonggu Sohn's Data Visualization, First Steps

This lesson turns everything so far upside down. Not one of the tricks below alters a number. And yet the reader walks away with a conclusion that is not true.

Switch the tricks on one at a time and see how far one dataset can be pushed.

Loading…

1. Truncated axes

You met this in lessons 4 and 5. Lift the baseline of a bar and a small difference becomes dramatic. The same happens with lines — narrow the y-range and noise looks like a trend.

2. Dual y-axes

Draw two different quantities in one panel with a y-axis on each side. Now the crossing point lands wherever you set the two scales. “The two curves meet here” means nothing at all — push one axis slightly and it disappears.

3. Cherry-picking

One representative image, one representative curve. Choose the best-looking one and the figure is dramatic no matter what the average says. Which is why good papers put the quantification of everything next to the representative picture — that is exactly why Fig. 4E pairs slide images with a box plot.

4. Double ink

3-D bars, shadows, perspective. One value carried simultaneously by length, area and volume. People read the most exaggerated of the three.

A 7% change. The middle panel makes it look like more than a tripling. Not one number was touched.

Python Runs in your browser

import matplotlib.pyplot as plt
import viz, numpy as np

vals = [100, 104, 103, 107]
labs = ["Q1", "Q2", "Q3", "Q4"]

fig, ax = plt.subplots(1, 3, figsize=(11, 3.2))
ax[0].bar(labs, vals, color=viz.PAL[0]); ax[0].set_ylim(0, 120)
ax[0].set_title("honest (starts at 0)", fontsize=9)

ax[1].bar(labs, vals, color=viz.PAL[1]); ax[1].set_ylim(98, 108)
ax[1].set_title("truncated - same numbers!", fontsize=9)

ax[2].plot(labs, vals, "-o", color=viz.PAL[2]); ax[2].set_ylim(99, 108)
ax[2].set_title("line, zoomed", fontsize=9)

for a in ax:
    viz.bare(a)
fig.tight_layout()
print("The total change is %.0f%%." % (100*(vals[-1]-vals[0])/vals[0]))

A sentence like “A overtakes B at this point” has no meaning on a dual-axis chart.

Python Runs in your browser

import matplotlib.pyplot as plt
import viz, numpy as np

# dual y-axis - move the crossing point by changing only the right-hand scale
x = np.arange(12)
a = 100 + x*3
b = 50 - x*1.2

fig, ax = plt.subplots(1, 2, figsize=(9.5, 3.2))
for j, (lo, hi) in enumerate([(30, 55), (10, 90)]):
    ax[j].plot(x, a, color=viz.PAL[0]); ax[j].set_ylabel("A", color=viz.PAL[0], fontsize=8)
    t = ax[j].twinx()
    t.plot(x, b, color=viz.PAL[1]); t.set_ylabel("B", color=viz.PAL[1], fontsize=8)
    t.set_ylim(lo, hi)
    ax[j].set_title("right axis %d~%d" % (lo, hi), fontsize=9)
fig.tight_layout()
print("Same data. Only the right-hand range changed - and the crossing point moved.")

🤔 Check yourself

What does the crossing point of two curves on a dual-axis chart mean?

🤔 Check yourself

A paper shows one representative histology image. What must accompany it?

🤔 Check yourself

Why are 3-D bar charts bad?