Joint Plots

A plain scatter plot answers "how do these two variables relate?" — but it stays silent on a second question that often matters just as much: "how is each variable distributed on its own?" A joint plot answers both at once.

sns.jointplot draws two numeric variables as a central bivariate plot, and tucks each variable's univariate distribution into the margins — one along the top, one down the right side. The center shows the joint relationship; the margins show the two marginals. You get the relationship and both distributions in one compact figure.

The minimal joint plot

By default the center is a scatter plot and the margins are histograms:

Read it in two directions. The center is the familiar scatter — bill length against bill depth. The top margin is the distribution of bill_length_mm alone; the right margin is the distribution of bill_depth_mm alone. Notice the top histogram looks distinctly two-humped — a clue there are subgroups hiding in this data that a bare scatter wouldn't have surfaced as clearly.

QuestionSelect one

In a jointplot, what is shown in the top and right margins?

Zoomed-in copies of the central scatter plot.

The univariate distribution of each variable — x along the top, y down the right.

The x-axis and y-axis labels, enlarged.

The residuals from a regression line.

Choosing the center with kind

The kind parameter swaps out what the central plot is, while the margins adapt to match. The options:

• kind="scatter" — the default; one dot per observation.
• kind="kde" — smooth density contours, like a topographic map of where points concentrate.
• kind="hist" — a 2-D histogram, binning the plane into colored rectangles.
• kind="hex" — hexagonal bins; the go-to fix for overplotting in dense data (more on this below).
• kind="reg" — a scatter with a regression line (and confidence band) drawn through it.

Here is kind="kde", which trades individual points for density contours — useful when you care about where the mass is rather than each observation:

Handling density with kind="hex"

A scatter's weakness is overplotting: with thousands of points the dots pile into a solid blob and you lose the very structure you came to see. A hexbin plot fixes this by dividing the plane into hexagonal bins and coloring each bin by how many points fall in it — darker means denser. No dot is ever hidden behind another, because you're counting, not stacking.

To show real density we need more points than the small built-ins offer, so here we draw a manageable random sample of the larger diamonds dataset:

A plain scatter of carat against price would be a dark smear in the bottom-left corner. The hexbin instead shows where diamonds actually concentrate — most are small and cheap, with a thinning tail toward larger, pricier stones — and the marginal histograms confirm both variables are heavily right-skewed.

QuestionSelect one

You plot 50,000 rows with jointplot(..., kind="scatter") and the center is one solid dark mass. Which kind most directly fixes the overplotting?

kind="reg"

kind="hex"

kind="scatter" with a larger marker size.

There is no way to show dense data in a joint plot.

Splitting groups with hue

Just like a scatter plot, a joint plot accepts a hue column. The center colors points by group, and — neatly — the margins split into one distribution per group, so you see how the categories differ on each axis:

That two-humped top margin from the very first plot now explains itself: the humps are different species. With hue the center resolves into clean clusters, and each margin shows three separate distributions — the same group-revealing payoff you saw with pair plots, focused on a single pair.

What a joint plot shows — and when to skip it

• Data it needs: two numeric columns, plus an optional categorical column for hue.
• What it highlights best: a single pair's relationship together with each variable's marginal distribution — skew, peaks, and outliers you'd miss in a bare scatter.
• When to prefer a plain scatter: when you only care about the relationship and the marginals are noise, a relplot/scatterplot is simpler and composes more flexibly (facets, custom Axes).
• When it breaks: large n with kind="scatter" overplots the center — switch to kind="hex" or kind="kde".

Your turn

Check your understanding

QuestionSelect one

What does sns.jointplot add compared with a plain scatter plot?

It facets the data into one panel per category.

It draws each variable's marginal distribution along the top and right of the central plot.

It computes a correlation matrix across all numeric columns.

It always fits and reports a regression model.

QuestionSelect one

Which kind turns the center of a joint plot into smooth density contours?

kind="hist"

kind="hex"

kind="kde"

kind="reg"

QuestionSelect one

When is a joint plot a better choice than a plain scatterplot?

Always — a joint plot is strictly superior and should replace every scatter.

When you need to place the chart on a specific matplotlib Axes alongside other plots.

When the marginal distributions of each variable matter, or you want a quick density view of a single pair.

When you want one panel per category.

You've now zoomed all the way in — from the whole-dataset pair-plot grid to a single, richly annotated pair of variables with its marginals. With relational and pairwise views in hand, the next chapters turn to the distribution plots (histograms, KDEs, ECDFs) that power those margins.