How to read a scatter plot and its clusters
A scatter plot places each case as a dot. Learn to read its axes, direction and clusters, to check outliers, and to see what a pattern can and cannot prove.
Each dot on a scatter plot is one case, such as one city, one student or one day, and its position carries two values at once. How far right it sits shows one variable, and how high it sits shows the other. A scatter plot answers a specific question: when one quantity is higher, does the other tend to be higher, lower or about the same? Reading one well means working through the parts in order: the axes and what the dots stand for, the overall pattern, any clusters, and the outliers.
Start with the axes. Check what each one measures, its units, the range it covers and whether it starts at zero. A plot of hours studied against exam scores tells you little if you do not know whether each dot is a student, a class average or a single exam sitting. Also look for a caption that states how many dots there are. A scatter of 12 dots and one of 1,200 dots can look alike at a glance, but the second pattern deserves far more confidence.
Direction, shape and strength
Describe the overall pattern in three parts. The direction is positive when the cloud slopes upward from left to right, meaning higher values on one axis go with higher values on the other, and negative when it slopes downward. The form is roughly linear when the dots cluster around a straight band, or curved when they follow a bend. The strength is how tightly the dots hug that pattern. A narrow band suggests a strong relationship, while a broad, fuzzy cloud suggests a weak one.
Two details of presentation can change how strong the pattern looks. Stretching one axis makes the same cloud look steeper or flatter. Also, if the plot covers only a narrow slice of one variable, the relationship can appear weaker than it is across the full range. Stretching and narrowing are both legitimate choices, but they change the visual impression. Compare the axis ranges with the range of the data before you judge the strength of the pattern.
The spread around the pattern tells you something as well. Where the dots sit close to the band, one variable predicts the other closely within this data set. Where they scatter widely, other factors are doing much of the work, and a single trend line gives only a rough guide to any individual case.
Clusters and what they suggest
A cluster is a group of dots that sits together, apart from the rest. Clusters often mark subgroups: two types of product, two regions, or a set of cases measured under different conditions. They matter because they suggest the data may not come from one population. If a plot shows two clusters, a single trend line drawn through both may describe neither group well.
This is where an overall pattern can mislead. Imagine a plot of two groups in which each group, on its own, rises from left to right, yet the groups sit at different heights so that the combined cloud looks flat. The combined cloud can even slope the opposite way from each cluster. This reversal is often called Simpson's paradox, and it means the pattern across everything can vanish or flip once the data are split by a grouping variable. When you see clusters, check whether color, shape or labels indicate a group, and ask whether the trend holds within each one.
Outliers before you judge the cloud
An outlier is a dot far from the rest. Before deciding what to do with it, ask three questions. Is it a data entry error, such as a value recorded in the wrong unit? Does it belong to a different group that should be shown separately? Or is it a real, extreme case that deserves attention in its own right? The answer determines whether the dot should be checked, separated or kept. An analysis that quietly drops outliers without saying so can change the conclusion, so look for a note on how they were handled.
Outliers also affect how a trend line looks. A single far-off point can tilt a fitted line, so the same cloud may appear to rise steeply with the outlier and to be nearly flat without it. If a chart includes a fitted line, check whether the line changes much when the most extreme dot is set aside. A line that depends on one point is a weak summary of the cloud.
What a pattern cannot prove
A clear pattern shows that two quantities vary together in the data shown, which statisticians call association. It does not show that one causes the other. Both could be driven by a third factor, or the arrow of influence could run the other way. A scatter plot also says nothing about cases outside the range it covers, so a trend that holds for small values may not hold for large ones. Treat the plot as a strong hint that deserves a closer look, and find the source's description of how the data were collected before drawing a causal conclusion.
When you meet a scatter plot in a report, try a short routine. Read the axes, decide whether the cloud rises or falls, and note how tight it is. Then check the clusters and the outliers, and ask whether either one changes that answer. The routine takes a minute, and it keeps a single striking dot from being mistaken for the whole story.