Skip to content
Bamboo Weekly

pandas plot.scatter

Two columns, one point per row.

Why does this one refuse to run without arguments, when .plot.line() and .plot.bar() are happy on their own? Because a scatter plot has no default. Line and bar both read the index for one axis and the columns for the other, so the shape tells them everything. A scatter plot has two axes and no index involvement at all — it needs to be told which column is horizontal and which is vertical, and there is no sensible guess.

So .plot.scatter(x='...', y='...') is the minimum, and every row of the frame becomes one point.

Official documentation: pandas.DataFrame.plot.scatter.

The arguments that earn their keep

df.plot.scatter(x='population', y='co2_per_capita')
df.plot.scatter(x='population', y='co2_per_capita', logx=True)   # skewed x
df.plot.scatter(x='a', y='b', alpha=0.3)     # see through overlapping points
df.plot.scatter(x='a', y='b', s='population')  # size by a third column
df.plot.scatter(x='a', y='b', c='continent')   # color by a third column
df.plot.scatter(x='a', y='b', figsize=(9, 7))

alpha is the one that rescues most real scatter plots. Anywhere past a few hundred points the marks pile up and a dense region looks identical to a moderately busy one; dropping opacity to 0.3 turns overlap back into information. logx and logy come next, because so many interesting quantities — population, income, company size — are spread across orders of magnitude, and on a linear axis every point but the outliers ends up in the corner.

A worked example, on real data

Population against emissions per person, one point per country:

import pandas as pd

url = 'https://raw.githubusercontent.com/owid/co2-data/master/owid-co2-data.csv'

df = pd.read_csv(url, usecols=['country', 'iso_code', 'year',
                               'co2_per_capita', 'population'])

(df
 .loc[(df['year'] == 2023) & (df['iso_code'].str.len() == 3)]
 .dropna(subset=['co2_per_capita', 'population'])
 .plot.scatter(x='population', y='co2_per_capita', logx=True, alpha=0.5))

That gives 213 points. The distribution behind them explains both extra arguments:

count    213.00
mean       4.61
std        5.51
min        0.06
25%        0.99
50%        3.14
75%        5.76
max       40.13

Per-capita emissions run from 0.06 to 40 tonnes, and population from tens of thousands to over a billion. On linear axes almost every country sits crushed against the left edge while China and India occupy the rest of the canvas, so logx=True is doing the real work here. alpha=0.5 handles the cluster of small, low-emitting countries that would otherwise be one solid blob.

The dropna is not optional. plot.scatter silently skips rows where either coordinate is missing, so without it you get a chart drawn from an unknown number of countries.

Three mistakes people make

Forgetting that it drops missing rows quietly. Line charts leave a visible gap; scatter plots just have fewer points, and nothing tells you how many. Run .isna().sum() on both columns, or dropna(subset=[...]) explicitly, so the point count is a decision rather than an accident.

Plotting on linear axes out of habit. If a column's maximum is a thousand times its median, the linear chart is a picture of the outliers and nothing else. Check with .describe() before you draw.

Reading a shape as a cause. A scatter plot is very good at suggesting relationships that turn out to be population size in disguise. If both axes scale with something else, corr on the residuals is a better next step than another chart.

Where it shows up in Bamboo Weekly

Bamboo Weekly #49: Campaign finance is the method at its plainest — df.plot.scatter(x='TTL_DISB', y='TTL_RECEIPTS'), money in against money out, one point per campaign. When the two axes are the same kind of quantity, the interesting thing is the diagonal.

Bamboo Weekly #60: Iceland and Bamboo Weekly #72: City travel both build the frame at some length before scattering it — cleaning a price column in one case, stacking and joining in the other. That is typical: the scatter call is one line and the reshape above it is ten.

Practice it

Work through a plot.scatter exercise, with instant feedback and no signup required: practice.lernerpython.com/bamboo-weekly/plot-scatter/

Go deeper

corr puts a number on what the scatter plot suggests. plot.hist is worth drawing first on each axis, to see whether a log scale is needed. dropna makes the missing rows explicit. For hover labels telling you which country a point is, plotly express scatter is the better tool.

More Pandas videos on Python and Pandas with Reuven Lerner.

Bamboo Weekly is the practice. If you want the structured version — full courses with downloadable Jupyter notebooks, plus live Pandas office hours when you get stuck — that is LernerPython+Data. A paid Bamboo Weekly subscription is included with it.

See it on real data

Below are the 9 Bamboo Weekly exercises that use plot.scatter on real-world data — try each one, then study the worked solution.

Part of the Pandas Methods Index. See also practice by skill.