2. The order your eyes read best — position, length, angle, area, colour · Dr. Seonggu Sohn's Data Visualization, First Steps

When a number becomes a picture, it is carried by one property your eye can read. That property is called a visual channel.

Channel	How the number rides	Where you meet it
Position	a coordinate on an axis	scatter · volcano · growth curve
Length	the length of a bar	bar chart
Angle	how far a wedge opens	pie chart
Area	the size of a circle	dot size in a dot plot · radius-scaled pies
Shade · hue	the colour of a cell	heatmap · cell types on a UMAP

What matters is that these five are not equally accurate.

The master-regulator scores from Fig. 2C of the paper, drawn through five channels. Which one lets you read the ranking?

Loading…

The numbers in the widget are real — the master-regulator scores the paper published in Fig. 2C.

DDX54 237 · ATP2A2 230 · RACGAP1 223 · PTPN11 221 · PCNA 218 · DIABLO 214 …

First place beats second by 7 points, about 3% of the total. Drawn as bars, that gap is visible. Drawn as a pie, it is not. And the paper's conclusion is “DDX54 is first” — so they had to pick a chart that shows 3%.

The same eight numbers. In the bars the ranking is immediate; in the circles it is vague; in the colours it is all but gone.

Python Runs in your browser

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

mr = viz.mr().sort_values("score", ascending=False).head(8)
g, s = list(mr.gene), list(mr.score)

fig, ax = plt.subplots(1, 3, figsize=(10.5, 3.4))

# 1) length - bars
ax[0].barh(range(8)[::-1], s, color=viz.PAL[0])
ax[0].set_yticks(range(8)[::-1]); ax[0].set_yticklabels(g, fontsize=7)
ax[0].set_title("length (bar)", fontsize=9); viz.bare(ax[0])

# 2) area - circles
ax[1].scatter(range(8), [0]*8, s=[(v/237*40)**2 for v in s], color=viz.PAL[1])
ax[1].set_xticks(range(8)); ax[1].set_xticklabels(g, rotation=90, fontsize=7)
ax[1].set_yticks([]); ax[1].set_title("area (circle)", fontsize=9); viz.bare(ax[1], ("bottom",))

# 3) shading - a one-row heatmap
ax[2].imshow(np.array(s).reshape(1, -1), cmap="Blues", aspect="auto")
ax[2].set_xticks(range(8)); ax[2].set_xticklabels(g, rotation=90, fontsize=7)
ax[2].set_yticks([]); ax[2].set_title("shade (heat)", fontsize=9)

fig.tight_layout()

And yet the dot plots in Figs. 5C and 7B load size and colour onto a single dot — two of the weakest channels at once. Why?

Because of this

40 genes × 12 cell types = 480 cells. As bars that is 480 bars. They do not fit on one screen.

So read it like this

You never read “how many times” off a dot plot. You read which cells are big and dark — the pattern, nothing else.

The weak channels are a trade: accuracy sold for volume. Whether that trade is fair is the subject of lesson 14.

🤔 Check yourself

In the Cleveland–McGill ranking, which channel is the most accurate?

🤔 Check yourself

You want the expression of 40 genes across 12 cell types on one screen. What fits best?

🤔 Check yourself

In Fig. 2C the gap between first (DDX54, 237) and second (ATP2A2, 230) is 3%. Which chart is worst at showing it?