15. Small multiples — putting four conditions side by side · Dr. Seonggu Sohn's Data Visualization, First Steps

The experiment in this paper is a 2×2.

	Isotype antibody	anti-PD1
WT Ddx54	control	immunotherapy only
Ddx54 KD	knockdown only	both

The paper's conclusion is that only the bottom-right cell works. So the four cells have to be drawn so they can be compared at a glance.

The tool for that is small multiples — the same chart repeated once per condition and laid out side by side.

Compare the four conditions overlaid against faceted. Then break the shared axis and see.

Loading…

The line sharex=True, sharey=True is the whole point of this figure. Without it the four panels cannot be compared.

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 = np.array(G["days"])
ymax = max(np.array(a["rows"]).max() for a in G["arms"])
tags = ["WT + isotype", "WT + anti-PD1", "KD + isotype", "KD + anti-PD1"]

fig, ax = plt.subplots(2, 2, figsize=(8.6, 5.4), sharex=True, sharey=True)
for i, arm in enumerate(G["arms"]):
    a = ax[i // 2][i % 2]
    R = np.array(arm["rows"])
    for row in R:
        a.plot(days, row, lw=.7, alpha=.4, color=viz.GREY)
    a.plot(days, R.mean(axis=0), lw=2.2, color=viz.PAL[i])
    a.set_title(tags[i], fontsize=9)
    a.set_ylim(0, ymax*1.05); viz.bare(a)
for a in ax[1]:
    a.set_xlabel("day", fontsize=8)
for a in ax[:, 0]:
    a.set_ylabel("volume (mm3)", fontsize=8)
fig.tight_layout()
print("(fake data, for shape only) Sharing the axes is what lets you compare the four heights directly.")

Overlay or facet — when do you use which?

Overlay

when there are four lines or fewer and the point is when they diverge. That is Fig. 4C — four curves in one panel

Facet

when points overlap, or when each condition needs its whole distribution shown. That is the density plots of Figs. 5K and 7C

The paper does both. Curves overlaid, densities faceted. Overlay while it still reads; split when it stops. The rule really is that simple.

🤔 Check yourself

You are laying four density plots side by side. What matters most?

Answer The point of laying them out is comparison. Different axis or colour ranges destroy that point. In matplotlib that means sharex/sharey and vmin/vmax.

🤔 Check yourself

Twelve lines overlaid in one panel are unreadable. What do you do?