Written and maintained by CASRAI Editorial Board
Last updated
Scatterplot interpretation means reading four things off the picture: the direction of the pattern, its form (straight or curved), its strength (how tightly the points follow that form), and any points that sit away from the rest. This guide walks through each step, works one example by hand, and shows why the picture should be read before any correlation number is quoted.
Definition in one paragraph
A scatterplot is a plot of the values of Y against the corresponding values of X. The NIST/SEMATECH e-Handbook of Statistical Methods describes its purpose as checking for a relationship: the vertical axis normally holds the response variable and the horizontal axis holds the variable suspected of being related to it. Each point is one observation, so a plot of 40 participants has 40 points. Interpreting it means describing the overall pattern in those points, using direction, form and strength, and then noting the points that deviate from it.
Reading direction, form and strength
Work through the plot in this order. The three descriptors are the standard vocabulary used in introductory statistics courses, such as the Open Learning Initiative material published as Concepts in Statistics.
Direction
If Y tends to rise as X rises, the relationship is positive. If Y tends to fall as X rises, it is negative. The relationship can also be neither: a shapeless cloud with no tilt shows no relationship between the two variables.
Form
Form is the shape the points follow. In a linear form the points scatter about a straight line; in a curvilinear form they scatter about a smooth curve. NIST lists quadratic, exponential and damped sinusoidal patterns as examples of non-linear relationships a scatterplot can reveal.
Strength
Strength describes how closely the points follow the form. Points hugging the line or curve indicate a strong relationship; a wide band indicates a weak one. A correlation coefficient puts a number on this for linear patterns only (see the correlation coefficient guide).
Outliers and spread
Outliers are points that deviate from the pattern of the relationship. Separately, ask whether the vertical spread of the points stays the same across X. NIST calls a plot where the variation in Y does not depend on X homoscedastic, and one where it does depend on X heteroscedastic. A band of points that widens or narrows along X is one sign of the second case; the residual plot guide explains why it matters once you fit a regression.
| What you see | How to read it |
|---|---|
| Points rise left to right around a line | Positive linear relationship |
| Points fall left to right around a line | Negative linear relationship |
| Points follow a smooth bend | Curvilinear form; a straight-line summary will misdescribe it |
| Shapeless cloud | No relationship evident |
| Band of points widens or narrows along X | Variation in Y depends on X (heteroscedastic) |
| One or more points far from the pattern | Outliers: check them before summarising |
Worked example with the arithmetic shown
Suppose eight students recorded weekly study hours (X) and a test score (Y). The values are illustrative numbers chosen for the calculation, not data from a study.
| Hours (X) | Score (Y) | X − 4.5 | Y − 63.25 | Product |
|---|---|---|---|---|
| 1 | 52 | −3.5 | −11.25 | 39.375 |
| 2 | 55 | −2.5 | −8.25 | 20.625 |
| 3 | 59 | −1.5 | −4.25 | 6.375 |
| 4 | 60 | −0.5 | −3.25 | 1.625 |
| 5 | 66 | 0.5 | 2.75 | 1.375 |
| 6 | 67 | 1.5 | 3.75 | 5.625 |
| 7 | 72 | 2.5 | 8.75 | 21.875 |
| 8 | 75 | 3.5 | 11.75 | 41.125 |
The means are 4.5 hours and 63.25 points. The products sum to 138.0. The squared X deviations sum to 42, and the squared Y deviations sum to 459.5. The Pearson correlation is then r = 138 / √(42 × 459.5) = 138 / √19,299 = 138 / 138.92 = 0.993.
Plotted, the points rise steadily and sit close to a straight line: positive direction, linear form, strong. Now add a ninth student who studied 9 hours and scored 30. The means become 5 hours and 59.56 points, the sum of products falls to 5, the squared X deviations sum to 60, and the squared Y deviations to 1,442.22. Now r = 5 / √(60 × 1,442.22) = 5 / 294.17 = 0.017. One point changed a near-perfect linear coefficient into a value close to zero, although eight of nine points still follow the same line. The scatterplot shows this at a glance; the coefficient alone does not.
The reverse problem also occurs. For X = −3 to 3 and Y = X², the points form a perfect U-shape, yet the sum of products is 0 and r = 0. A coefficient of zero does not mean “no relationship”; it means no straight-line relationship.
Assumptions and common mistakes
- Reading a coefficient without the picture. As the example shows, one outlier or one curve can make r misleading. Plot first.
- Assuming a straight line. Check the form before choosing a linear summary.
- Swapping the axes carelessly. NIST puts the response on the vertical axis and the suspected explanatory variable on the horizontal axis; keep to that convention so readers interpret direction correctly.
- Claiming cause. NIST states that association does not imply causality: a scatterplot can show that two variables are related, not that one produces the other. See correlation vs causation.
How to report it
Describe the pattern in words, give the sample size, and add the coefficient only when the form is linear. For the eight-student example:
A scatterplot of study hours and test score (n = 8) showed a positive, approximately linear, strong relationship, r = 0.99. No points departed from the pattern.
Format numbers as your journal or style guide requires. Say “associated with” rather than “caused by”, and state the number of outliers and how they were handled. For figure design, see the guide on designing effective scientific figures.
How to calculate and plot it in Excel, R or Python
Excel
Select both columns and insert a Scatter chart, then compute the coefficient with this formula (data in A2:A9 and B2:B9):
=CORREL(A2:A9, B2:B9)
R
x = 1:8 y = c(52, 55, 59, 60, 66, 67, 72, 75) plot(x, y, xlab = "Study hours", ylab = "Test score") cor(x, y) # 0.9933715
More detail is in scatterplots in R.
Python
import numpy as np
import matplotlib.pyplot as plt
x = np.arange(1, 9)
y = np.array([52, 55, 59, 60, 66, 67, 72, 75])
plt.scatter(x, y)
plt.xlabel("Study hours")
plt.ylabel("Test score")
print(np.corrcoef(x, y)[0, 1]) # 0.99337...
plt.show()
See scatter plot in Python for options such as colour by group.
FAQ
What does a scatterplot show?
It shows whether two quantitative variables are related, whether the relationship is linear or non-linear, whether the variation in Y changes with X, and whether outliers are present.
How do you describe a scatterplot?
State the direction (positive, negative or neither), the form (linear or curvilinear), the strength (how closely points follow the form), and any outliers.
Does a strong pattern mean X causes Y?
No. A scatterplot can show association but cannot prove cause, as the NIST handbook notes.
What if the points curve but r is near zero?
Report the curved form. A near-zero r only rules out a straight-line relationship, as the Y = X² example above shows.
What is the difference between a scatterplot and a residual plot?
A scatterplot shows the raw Y against X. A residual plot shows what is left after a fitted line is removed; see regression analysis and the residual plot guide above.
Sources
- NIST/SEMATECH e-Handbook of Statistical Methods, section 1.3.3.26, “Scatter Plot”: https://www.itl.nist.gov/div898/handbook/eda/section3/scatterp.htm (checked 2026-10-29).
- Concepts in Statistics (Lumen Learning, content from the Open Learning Initiative, CC BY 4.0), “Scatterplots (2 of 5)”: https://stats.libretexts.org/Courses/Lumen_Learning/Concepts_in_Statistics_(Lumen)/03%3A_Examining_Relationships-_Quantitative_Data/3.15%3A_Scatterplots_(2_of_5) (checked 2026-10-29).
- Worked-example figures: calculated from the tables on this page.
