A scatter plot is a chart that shows each observation as a point positioned by two numeric variables, one on the horizontal x-axis and one on the vertical y-axis. In data visualization, it is the primary chart for examining the relationship between two quantitative variables: whether they move together, how strongly, in what shape, and which observations deviate from the pattern.
Unlike a bar chart or a line chart, a scatter plot does not summarize data before drawing it. Every point is one record, so the chart exposes distribution, clusters, gaps and outliers that averages hide. Additional variables can be encoded through color, size or shape, and the chart is often extended with a fitted trend line to support, but not replace, visual judgment.
How a scatter plot encodes data
A scatter plot relies on position along a common scale, the most accurate visual encoding the human eye can read. Each point’s coordinates come directly from its values, which makes the chart precise for comparison and quick to scan for patterns.
| Channel | Encodes | Practical guidance |
|---|---|---|
| X position | Explanatory or input variable | Place the presumed cause or driver here |
| Y position | Response or outcome variable | Place the measured effect here |
| Color | Category or a third numeric variable | Use at most about six distinct categories |
| Size | Magnitude of a third variable | Scale by area, not radius; becomes a bubble chart |
| Shape | Secondary category | Use sparingly; shapes are harder to distinguish than color |
| Opacity | Density of overlapping points | Semi-transparent points reveal crowded regions |
The choice of axes carries meaning. By convention, the independent variable goes on the x-axis and the dependent variable on the y-axis. When neither variable drives the other, the order is arbitrary, but it should stay consistent across related charts.
Patterns a scatter plot reveals
Reading a scatter plot means looking for direction, form, strength and exceptions. A correlation coefficient compresses these into one number, but only the plot shows whether that number is meaningful.
| Pattern | What it looks like | What it suggests |
|---|---|---|
| Positive linear | Points rise from lower left to upper right | Variables increase together |
| Negative linear | Points fall from upper left to lower right | One variable decreases as the other increases |
| Nonlinear | Curved band, such as saturation or a U shape | A straight-line model would mislead |
| No relationship | Shapeless cloud | Little linear or monotonic association |
| Clusters | Separate groups of points | Hidden categories or segments |
| Fan shape | Spread grows with x | Variance changes with level, often multiplicative effects |
| Outliers | Isolated points far from the pattern | Errors, special cases or valuable exceptions |
Very different datasets can share identical means, variances and correlation yet look completely different when plotted. This is the main argument for always plotting paired numeric data before relying on summary statistics.
Scatter plots compared with other chart types
The scatter plot answers a specific question: how do two numeric variables relate across individual observations. Other charts in the visualization toolbox answer different questions and should not be substituted for it.
| Chart | Best question | When to prefer it over a scatter plot |
|---|---|---|
| Scatter plot | How do two numeric variables relate | Not applicable, the reference chart |
| Line chart | How does a value change over ordered time | X is time and points form one connected series |
| Bar chart | How do categories compare | One variable is categorical |
| Histogram | How is one variable distributed | Only one numeric variable matters |
| Heatmap or 2D histogram | Where are points densest | Tens of thousands of points overlap |
| Scatter plot matrix | How do several variables relate pairwise | Exploring four or more numeric variables |
Design choices that improve readability
Good scatter plots come from a few deliberate design decisions. Default settings in charting tools often produce cluttered or misleading results, especially with large datasets.
- Handle overplotting: with more than a few thousand points, use transparency, smaller markers, jitter for discrete values or a density display.
- Choose axis scales carefully: use a logarithmic scale when values span several orders of magnitude, such as incomes or city populations.
- Do not force zero on both axes: unlike bar charts, scatter plots do not encode length, so axes may start where the data begin.
- Add a trend line only with a reason: a linear fit on curved data is misleading; consider a smoothed curve such as LOESS.
- Label selected points, not all of them, to direct attention to meaningful outliers.
- Keep aspect ratio sensible: stretching one axis exaggerates or flattens slopes.
Misreadings and limitations
The most common misreading of a scatter plot is treating visible association as proof of causation. A third variable can drive both axes, and the direction of influence may be the reverse of what the axis order implies.
| Problem | Effect | Remedy |
|---|---|---|
| Confounding variable | False impression of cause and effect | Color points by the suspected confounder |
| Overplotting | Dense regions look like a solid blob | Transparency, sampling or density plots |
| Mixed groups | Overall trend reverses within groups | Split by group or use facets |
| Restricted range | Real relationship appears weak | Check whether data cover the full range |
| Discrete values | Points stack invisibly | Apply small random jitter |
Mixing groups deserves particular attention. When subgroups with different baselines are plotted together, the overall trend can point in the opposite direction from the trend inside every group. Faceting or color by group usually reveals this immediately.
In everyday language
A scatter plot is a chart made of dots, one dot for each thing being measured. Think of a class where every student is a dot. The dot’s left-right position shows hours of study. Its up-down position shows the test score.
If the dots form a band rising to the right, more study goes with higher scores. If they form a random cloud, there is no clear link. The chart shows the pattern and the exceptions at a glance, without hiding anyone in an average.
Seen in the real world
A regional chain of 140 convenience stores wants to know whether store size explains sales. An analyst plots floor area on the x-axis and annual revenue on the y-axis, one point per store, with color showing location type: city center, suburban or highway.
The overall cloud shows only a weak positive trend. Colored by location, the picture changes: city-center stores sit on a steep band with high revenue per square meter, highway stores form a flat cluster, and a few suburban stores fall far below their peers.
Management drops a plan to enlarge highway stores, which the chart shows would add little revenue. Instead, it investigates the five underperforming suburban outliers and finds poor visibility and parking. The scatter plot turned a single average into specific, actionable decisions.