Scatter Plot

Plotting two measures together to see whether they move in step.

Also known as: scatterplot, scatter graph, scatter chart

A scatter plot puts one measure on the horizontal axis and another on the vertical axis, with one dot per row. It answers “do these two numbers move together”, and it answers with a shape rather than with a number.

What to look for

A rising or falling band: the two measures move together.
A cloud with no direction: they do not.
Clusters: more than one population is mixed into the same axes.
Curves: the relationship is real but not a straight line, and a correlation number would have hidden that.
Points far from everything else: see outliers. Often the most informative dot on the plot.

The number and its limits

If you see a band, the number that describes its tightness is the correlation coefficient. It is not the whole answer, and the reason is correlation versus causation: a scatter plot showing that taller customers spend more has not shown that height causes spending. Both may track age.

When you suspect a third factor, colour the dots by it. A band that splits into two clean coloured clusters has just told you that your single relationship was two.

Classic mistakes

Overplotting. Many rows land on the same spot and you see one dot where there are two hundred. Use transparency, or bin the pairs into a heatmap of the density.
Scaling both axes so a strong relationship looks weak, or a weak one looks strong. The axis range is a choice — see axis truncation — and while zooming is legitimate here, say what you did.
Categorical colour for eight categories, which turns the plot into confetti and tests the reader’s colour memory instead of showing the data.