Reading a Scatterplot

Introduction

A scatterplot turns paired measurements into a picture. Its horizontal and vertical positions let you inspect how a quantitative response varies across observed values of a quantitative predictor. To read the graph accurately, first identify what each axis measures; only then can a point or the whole cloud be interpreted in context.

A scatterplot is not just a collection of dots. Each dot represents a particular observational unit and preserves the pairing between that unit’s predictor and response values. The overall arrangement can reveal whether responses tend to differ across predictor values, but the plot alone does not explain why they differ or establish that one variable causes the other.

What the Axes and Points Represent

The axis label tells you which variable is shown, and its units tell you how to interpret a coordinate. For example, if the horizontal axis is labeled “Commute distance (km)” and the vertical axis “Commute time (minutes),” a point at \((8,25)\) represents one recorded commute that was 8 km long and took 25 minutes. It does not represent two separate units: the pairing belongs to the same commute.

Read a point by tracing its horizontal and vertical positions to the axes. The horizontal coordinate gives the predictor value; the vertical coordinate gives the response value. If a point falls between labeled tick marks, its value may need to be estimated from the scale. Do not report more precision than the display supports.

The axes describe the variables, not necessarily the people or objects in the data. A point might represent one student, one neighborhood, one day, or one manufactured item, depending on how the observations were collected. Check the description of the data before naming the observational unit. If several units have identical coordinates, their marks can overlap, so the visible number of dots may not equal the number of observations.

Reading the Point Cloud

After reading individual points, look across the whole cloud. Ask how the observed response values are distributed at lower, middle, and higher predictor values. Are the responses mostly similar throughout the horizontal range, or do their positions change as the predictor changes? Is there a visible pattern in the cloud, or does it look scattered without an obvious structure?

A useful way to keep this description grounded is to compare vertical positions at different horizontal locations. At a given predictor value, the plot shows the response values recorded for observations at or near that value. Moving across the graph lets you see whether the observed responses tend to shift, remain in a similar range, or change in a more complex way. A scatterplot displays observed values; it does not automatically provide a predicted or average response at every possible predictor value.

If the cloud has a clear overall upward tendency, higher observed predictor values tend to occur with higher observed response values in the plotted data. A downward tendency means higher predictor values tend to occur with lower observed responses. These descriptions summarize a broad tendency, not a rule that every point must follow. Points can sit above or below the overall pattern.

Do not decide that there is no association merely because the cloud has no consistent upward or downward tendency. A curved pattern, such as a U-shape, can show a clear relationship even though its response first decreases and then increases. Look across the full cloud for a non-linear pattern before concluding that there is no clear association. A more detailed vocabulary for describing direction, form, and strength is useful, but the first task is simply to report what the plotted observations visibly do.

Check the Full Pattern Before Summarizing

A short verbal summary should preserve the visible structure of the cloud rather than force it into a simple upward-or-downward description. One quick check is to consider whether a single broad direction describes the points across the entire horizontal range. If the cloud bends or reverses direction, say so. If there is no obvious structure, use cautious wording such as “no clear pattern is visible in these plotted observations,” rather than claiming that the variables can never be associated.

This check matters because summaries can hide important differences. For example, a response may be low at both ends of the predictor range and higher in the middle. Calling such a cloud “flat” because its endpoints are at similar heights would miss the pattern between them. A scatterplot should be read from left to right and across the whole cloud, not just by comparing the first and last points.

What a Scatterplot Can and Cannot Establish

A scatterplot is a descriptive display. It can show the observed pairs, make broad patterns easier to see, and reveal whether a simple summary might overlook visible structure. It cannot by itself establish a causal effect. A relationship in the plot may reflect other variables, the way observations were selected, or chance variation. Even when a study deliberately varies a predictor, conclusions about cause depend on the study design, not just on the visual appearance of the dots.

The plotted range also matters to interpretation: the cloud describes the predictor values actually shown. A pattern among observations in one range does not guarantee the same pattern outside that range. Likewise, a handful of points may suggest a tendency, but it provides limited evidence about how consistently that tendency would appear in a wider set of observations. Describe what is visible without extending the claim beyond the data.

When giving a verbal reading, include the variable names and units where they help clarify the meaning. “The points go up” is vague. “In these observations, higher commute distances generally occur with longer commute times” identifies both variables and states the visible tendency. A careful description also leaves room for exceptions and avoids implying that the plot proves an explanation.

Common Mistakes

• Swapping the variables. Read the axis labels before interpreting a point. The horizontal coordinate is the predictor, and the vertical coordinate is the response under the usual regression convention.
• Separating coordinates that belong together. A point at \((x_i,y_i)\) represents both values for one observational unit, not an arbitrary pairing of values from different units.
• Calling every pattern with no overall slope “no association.” A cloud can have a clear curved pattern without a consistent upward or downward tendency. Inspect the whole cloud for non-linear structure first.
• Turning an observed tendency into a universal rule. A general upward tendency does not mean every point lies higher than the point before it, and it does not supply an exact response for every predictor value.
• Claiming causation from the picture alone. A scatterplot can display an association, but the graph alone cannot show that changes in the predictor caused changes in the response.