Scatter plots exist to show how 2 numeric columns move together, one dot per row.
They are the visual partner to correlation, showing the shape that the single correlation number summarises.
An upward cloud means the columns rise together, and a downward cloud means one falls as the other rises.
A tight cloud means a strong relationship, while a round, shapeless cloud means little or none.
Curves show up clearly, which matters because correlation measures straight-line relationships only.
Separate clusters can show hidden groups, such as flats and houses forming 2 clouds in a price plot.
Colour or shape can add a third column, such as marking each dot by region.
Dot size can add a fourth, which makes a bubble chart, though sizes are hard to compare.
Overplotting is the main problem with large data: thousands of dots merge into a solid blob.
Transparency, smaller dots or a grid of counts shows where the dots are densest.
Adding a straight trend line helps the eye, but it can suggest a stronger relationship than the cloud shows.
A scatter plot shows that 2 things move together, never on its own that one causes the other.
In practice, plotting each pair of important columns early avoids trusting a correlation that hides a curve.
A grid of scatter plots for every pair of columns gives a quick overview of a whole table.
The usual mistake is drawing a line through the dots in row order, which turns a scatter plot into a meaningless zigzag.