4. Point, line, bar — the promise each mark makes · Dr. Seonggu Sohn's Data Visualization, First Steps

The body of a chart is only ever three things — points, lines and bars. A UMAP is points. A volcano plot is points. A survival curve is a line. A heatmap is a very wide bar.

But these are not merely shapes. They are promises made to the reader.

•

Point — “one of these happened here”

one observation, one dot. It says nothing at all about the space between

╱

Line — “the space between is filled”

it claims values exist between two points. Only for time, rank or a continuum

▮

Bar — “this much, from zero”

the length is the value. Which is why it must start at zero

Redraw the same data as points, lines and bars. Then lift the bar baseline and watch.

Loading…

The three marks can also be combined — most figures in the paper do.

Combination	What it becomes	In the paper
points + line	growth curve · enrichment curve	Figs. 4C · 6B
bars + points	a mean bar with the individuals scattered on it	Fig. 4B
points + box	distribution and individuals together	Fig. 4E
points alone	volcano · UMAP · dot plot	Figs. 3B · 5B · 5C

There is one rule: an added mark must add information. Scattering points over a bar adds the sample size and the spread, so it is good. Adding a drop shadow to a bar adds nothing, so it is bad.

The left panel is honest but hard to read; the middle one is readable but asserts unmeasured points; the bars on the right threw the whole process away.

Python Runs in your browser

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

F = json.load(open("viz_facts.json"))
g = F["growth"]
days = g["days"]
arms = [(a["name"], np.array(a["rows"]).mean(axis=0)) for a in g["arms"]]

fig, ax = plt.subplots(1, 3, figsize=(11, 3.3))
labs = ["WT+iso", "WT+aPD1", "KD+iso", "KD+aPD1"]

# points only
for i, (_, v) in enumerate(arms):
    ax[0].plot(days, v, "o", ms=4, color=viz.PAL[i])
ax[0].set_title("points only", fontsize=9)

# points + line
for i, (_, v) in enumerate(arms):
    ax[1].plot(days, v, "-o", ms=4, lw=1.6, color=viz.PAL[i], label=labs[i])
ax[1].set_title("points + line", fontsize=9)
ax[1].legend(fontsize=6.5, frameon=False)

# bars (last day only)
last = [v[-1] for _, v in arms]
ax[2].bar(labs, last, color=viz.PAL[:4])
ax[2].set_title("bars (day 21 only)", fontsize=9)
ax[2].tick_params(axis="x", rotation=45, labelsize=7)

for a in ax:
    viz.bare(a)
    a.set_xlabel("day" if a is not ax[2] else "", fontsize=8)
ax[0].set_ylabel("tumour volume (mm3)", fontsize=8)
fig.tight_layout()
print("This growth data is fake, for shape only - the paper's raw values were never published.")

Which of the three is right? It depends on the question.

The question	The mark
When did the arms diverge?	points + line
How big is the final difference?	bars (or a box with points on it)
How much do individuals vary?	one thin line per animal (spaghetti)

The paper puts bars in Fig. 4B and lines in Fig. 4C, side by side. Two questions, two answers.

🤔 Check yourself

Mean TIL for thirteen tumour types is plotted as dots joined by a line. What is wrong?

🤔 Check yourself

Two groups have means of 40 and 44. What happens if the bar y-axis starts at 38?