Scatter Plot
Also known as: scatterplot, scatter diagram, XY plot, correlation plot, scatter chart
A scatter plot (also called a scatterplot or scatter diagram) is a data visualization that places each data point at coordinates defined by two numerical variables -- one on the x-axis and one on the y-axis.
The resulting pattern of points reveals the relationship between the two variables: positive correlation (points trend from lower-left to upper-right), negative correlation (upper-left to lower-right), no correlation (randomly scattered), or more complex non-linear relationships. Scatter plots also reveal outliers (points isolated from the main cluster), clusters (subgroups of points), and heteroscedasticity (increasing spread of points at higher values). They are among the most fundamental exploratory data visualization tools.
Scatter Plot, explained at three levels
At home
A scatter plot shows the relationship between two variables by plotting each observation as a dot. If you plotted hours studied (x-axis) vs. test scores (y-axis) for 50 students, you would place each student as a dot at their specific (hours studied, test score) combination. If students who studied more tended to score higher, the dots would form a pattern trending upward from left to right -- showing a positive relationship. If the dots were scattered randomly with no pattern, the two variables would not be related. Scatter plots let you see relationships that a table of numbers would hide.
At work
Scatter plots are used in business to visualize relationships between metrics: customer satisfaction vs. purchase frequency, advertising spend vs. sales revenue, employee tenure vs. productivity. Before running any statistical analysis (correlation, regression), a scatter plot lets you see the raw relationship: is it linear or curved? Are there outliers pulling the analysis? Are there distinct clusters suggesting different customer types? The scatter plot is an exploratory tool that informs which analytical methods are appropriate and flags potential problems in the data before they affect model results.
At home
Reading scatter plot patterns: positive linear correlation (dots cluster around an upward-sloping line); negative linear correlation (downward-sloping line); no correlation (random scatter with no directional pattern); non-linear relationship (curved patterns -- a U-shape, S-curve, or other non-linear form); heteroscedasticity (the scatter of points increases at higher x-values, visible as a funnel shape widening to the right); outliers (isolated points far from the main cluster); clusters (two or more distinct groups of points). Each pattern has different implications for the statistical methods that should be applied and the business interpretation.
At work
Adding dimensions to scatter plots: a third variable can be encoded as dot color (categorical -- different colors for different customer segments) or dot size (numerical -- larger dots for higher revenue customers). This produces a three-variable scatter plot in two dimensions. Adding a fourth variable through dot shape is possible but makes the chart harder to read. Scatter plot matrices (pairs plot) show scatter plots for all pairs of variables in a dataset simultaneously, enabling rapid exploration of many relationships at once. Both approaches are standard in initial data exploration in data science work.
At home
Overplotting solutions for large datasets: when datasets have thousands or millions of points, scatter plots become a solid mass of overlapping dots that obscures the underlying pattern (overplotting). Solutions: transparency/alpha blending (making each point semi-transparent so overlapping regions appear darker); binning (dividing the plot area into grid cells and coloring by point density -- creating a 2D density histogram); hexbin plots (hexagonal binning that shows density naturally); contour lines (showing iso-density curves). Each solution reveals different aspects of the distribution: density plots show where points are most concentrated; contour lines show the shape of the density.
At work
Anscombe's quartet as a warning about summary statistics without visualization: Francis Anscombe constructed four datasets in 1973 that have identical statistical properties (same mean, variance, correlation coefficient, and regression line) but produce dramatically different scatter plots -- one linear, one curved, one with an outlier, one with a leverage point. Anscombe's quartet demonstrates that statistical summaries can completely miss the actual structure of data and that visualization is not optional for rigorous analysis. The lesson -- always plot your data before calculating statistics about it -- is a foundational principle in data analysis taught through Anscombe's example.
Real-world examples
At home
Carlos was analyzing whether his weekly exercise minutes correlated with his mood ratings (both tracked in an app). He made a scatter plot with exercise minutes on the x-axis and mood (1-10) on the y-axis. He could see: a moderate positive relationship (more exercise generally associated with better mood), but two clusters -- weeks below 30 minutes of exercise clustered around mood 4-5, and weeks above 60 minutes clustered around mood 7-8. The scatter plot revealed a non-linear threshold effect: a minimum exercise level mattered more than incremental increases above it.
At work
Aisha was evaluating regional sales performance for a retail chain. She plotted stores on a scatter plot with foot traffic (x-axis) vs. revenue per visitor (y-axis). She found most stores clustered in a moderate-traffic, moderate-revenue zone. But two outlier stores caught her eye: one with very high foot traffic but very low revenue per visitor (suggesting a location with poor conversion -- lots of browsers, few buyers) and one with low foot traffic but very high revenue per visitor (a smaller store converting at very high rates). These outliers drove a store operations investigation that revealed why some stores were outperforming in efficiency.
Common confusions
- Many people assume a tight, linear pattern in a scatter plot means one variable causes the other — Scatter plots reveal correlation (the strength and direction of association between variables), not causation. A tight linear pattern in a scatter plot shows a strong association, not that one variable causes changes in the other. Both might be caused by a third variable (confounding), one might predict the other without mechanically causing it, or the relationship might be coincidental. Correlation analysis of scatter plots is descriptive; establishing causation requires experimental or quasi-experimental methods regardless of how tight the scatter plot association appears.
- Some assume that a scatter plot showing no linear pattern means the two variables are unrelated — Scatter plots can show non-linear relationships that the linear correlation coefficient (r) does not capture. A perfect U-shaped relationship (y increases then decreases as x increases) has a correlation coefficient of 0 (no linear relationship) but a clear non-linear pattern. Temperature and ice cream sales have an approximately linear positive relationship up to very hot temperatures, where both peak and the relationship may change. Always visually inspect the scatter plot rather than only computing the correlation coefficient -- non-linear relationships are common and important.
For analysts
When delivering data analyses to clients, scatter plots are often the most effective visualization for demonstrating relationships between two business metrics. The practical considerations: label axes clearly with meaningful business names (not variable codes); annotate interesting points (outliers, specific stores or customers) with labels; add a trend line (regression line) to make the direction and strength of relationship visually explicit; and note the sample size (a scatter plot with 15 points tells a different story than one with 1,500 points). For presentations, scatter plots work best when pre-identified clusters or outliers are highlighted and narrated -- the audience needs guidance on where to look in the visual to understand your point.
History & origin
Francis Galton popularized scatter plots in the 1870s-1880s through his work on correlation and regression, using them to visualize the relationship between parents' and children's heights. Karl Pearson formalized the correlation coefficient as a mathematical measure of the relationship visible in scatter plots. Scatter plots became standard in scientific and statistical publications through the 20th century. Modern software (Excel, Tableau, ggplot2, Matplotlib, Seaborn) made generating scatter plots trivial, democratizing their use. The scatter matrix (multiple scatter plots of all pairs of variables in a dataset) is a standard exploratory visualization in data science for understanding multivariate relationships before modeling.