3. The kind of data decides half the chart for you · Dr. Seonggu Sohn's Data Visualization, First Steps

“Which graph should I use?” is usually the wrong question. Ask first: what kind is each column of my data? Fix that and the shortlist drops to three or four.

Kind	Example (in this paper)	Charts available
Categorical (names)	tumour types LUAD·LUSC·STAD / cell types T·B·NK	bars · dot plot · colour
Ordinal	TMB-L → TMB-H / stage I·II·III	bars (in order) · box
Count	616 mutations / a count of 802	histogram · scatter
Proportion	3.03% TIL infiltration / cell composition	bars · box · (pie)
Time	day 0·3·6·…·21 after implantation	line · step
Coordinate	cell position in tissue / UMAP location	scatter · density · map

Pick a dataset, then swap the chart. Mismatched pairs raise a warning.

Loading…

Only group is categorical; the rest are quantities. Writing down the kind of each column first makes the chart choice easy.

Python Runs in your browser

import viz, pandas as pd

m = viz.meta()
print("Lung adenocarcinoma sample table - each column is a different kind")
print()
for c in m.columns:
    v = m[c].dropna()
    num = pd.api.types.is_numeric_dtype(v)
    kind = ("count" if v.max() > 50 else "proportion") if num else "categorical"
    span = ("%.2f ~ %.2f" % (v.min(), v.max())) if num else ", ".join(sorted(set(map(str, v))))
    print("  %-6s %-11s n=%3d   %s" % (c, kind, len(v), span))

The same column can change kind depending on what you are asking. The paper's TMB is the clean example.

As a quantity

how many per Mb → histogram. You see how skewed the distribution is

As an order

TMB-L / TMB-H split at 10 → two bars. That is Fig. 1B of the paper

The paper did both. The right half of Fig. 1B is the quantity (a box plot), the left half is the ordering (proportion bars). Drawing one column twice is not waste, it is honesty — they did not hide the original distribution after drawing a threshold through it.

🤔 Check yourself

You want to show mean TIL infiltration for thirteen tumour types. What fits best?

🤔 Check yourself

Tumour volume is measured on days 0, 3, 6, 9, 12, 15, 18 and 21. What should the x-axis do?