Correlation coefficient (r) explained

By Jude Wallis · Published

The correlation coefficient rr is a number from −1-1 to 11 that measures the direction and strength of the linear association between two quantitative variables. Values near 11 or −1-1 mean the points fall close to a line. Values near 0 mean a weak linear association.

AP Statistics: Unit 5 (topics 5.2 Correlation). Correlation is Unit 5, topic 5.2 in the Fall 2026 AP Statistics course, where r describes the direction and strength of a linear association between two quantitative variables.

What the correlation coefficient measures

The correlation coefficient, written rr, summarizes a scatterplot of two quantitative variables in one number. It reports two things about the linear pattern:

  • Direction. A positive rr means that as xx increases, yy tends to increase. A negative rr means that as xx increases, yy tends to decrease.
  • Strength. The closer rr is to 11 or −1-1, the more tightly the points cluster around a straight line. The closer rr is to 0, the weaker the linear association.
Value of rrWhat the scatterplot looks like
r=1r = 1Every point on a line with positive slope
rr near 11Points close to a rising line
rr near 00Little or no linear pattern
rr near −1-1Points close to a falling line
r=−1r = -1Every point on a line with negative slope

Words like strong, moderate, and weak are judgments. There are no official cutoffs, so describe strength from the scatterplot and the value of rr together.

The correlation coefficient formula

r=1n−1∑(xi−xˉsx)(yi−yˉsy)r = \frac{1}{n-1} \sum \left( \frac{x_i - \bar{x}}{s_x} \right) \left( \frac{y_i - \bar{y}}{s_y} \right)
  • nn is the number of data points, each a pair (xi,yi)(x_i, y_i).
  • xˉ\bar{x} ("x-bar") and yˉ\bar{y} ("y-bar") are the sample means of the two variables.
  • sxs_x and sys_y are the sample standard deviations of the two variables.
  • xi−xˉsx\frac{x_i - \bar{x}}{s_x} is the z-score of xix_i, and yi−yˉsy\frac{y_i - \bar{y}}{s_y} is the z-score of yiy_i.

So rr is close to an average of the products of the z-scores. A point that is above the mean in both variables, or below the mean in both, has a positive product and pushes rr up. A point that is above the mean in one and below in the other has a negative product and pulls rr down.

In practice technology computes rr. Working one example by hand shows why the properties below hold. A longer hand calculation is in how to find r by hand.

When to use correlation

Use rr when all three of these hold:

  1. Both variables are quantitative. Correlation between a categorical variable such as eye color and anything else has no meaning.
  2. The scatterplot shows a roughly linear form. A strong curved pattern can produce an rr near 0, and a clearly curved pattern can also produce an rr near 1.
  3. There are no outliers distorting the pattern, or you have checked their effect. Like the mean and standard deviation it is built from, rr is nonresistant.

Always look at the scatterplot first. The number rr cannot tell you whether a line is the right model. A residual plot can.

How to interpret r in context

An interpretation states strength, direction, the word linear, and both variables in context:

"There is a strong, positive, linear association between hours studied and quiz score."

Two cautions on wording:

  • Say "association", not "causes". Correlation alone never shows that changes in xx cause changes in yy. See correlation vs causation.
  • rr is not a percentage. An rr of 0.96 does not mean 96% of anything. The percentage interpretation belongs to r2r^2, the coefficient of determination: about 92.4% of the variation in quiz score in the worked example is explained by the linear relationship with hours studied, since 0.96112≈0.9240.9611^2 \approx 0.924. See r vs r-squared.

Properties of r

  • Always between −1-1 and 11, inclusive.
  • No units. The z-scores in the formula have no units, so rr has none either.
  • Unchanged by a change of units. Converting hours to minutes, or points to percent, leaves rr exactly the same. In the worked example, measuring study time in minutes still gives r≈0.961r \approx 0.961.
  • Symmetric. Swapping which variable is xx and which is yy does not change rr.
  • Same sign as the slope of the least-squares regression line. The slope is b=r⋅sysxb = r \cdot \frac{s_y}{s_x}, and standard deviations are positive.
  • Measures linear association only. It says nothing reliable about curved relationships.
  • Nonresistant. One unusual point can raise or lower rr noticeably. Try moving a point in the regression and influential point interactive.

Common mistakes

  1. Claiming causation. A strong correlation in observational data can come from a third variable related to both.
  2. Calling the form linear because rr is large. A high rr does not prove a linear model is appropriate. Check the scatterplot and the residual plot.
  3. Reading r=0r = 0 as "no relationship". It means no linear association. A clear curved relationship can have rr near 0. See does a correlation of zero mean no relationship.
  4. Confusing rr with the slope. A steep line can have a small rr and a shallow line can have a large one. Slope has units and rr does not.
  5. Forgetting the sign when recovering rr from r2r^2. Take the square root, then attach the sign of the slope.
  6. Leaving out "linear" or the context in the interpretation sentence.

The correlation coefficient in AP Statistics

Correlation is Unit 5, topic 5.2 in the Fall 2026 AP Statistics course. Unit 5, Regression Analysis, carries 10-20% of the multiple-choice section.

Be ready to estimate and interpret rr from a scatterplot, find rr from regression output (often as the square root of r2r^2 with the sign of the slope), explain what happens to rr when units change or a point is added or removed, and separate association from causation.

Compute rr with the regression calculator, then work the correlation and residuals practice set and the scatterplots practice set.

Finding r for hours studied and quiz score

Six students reported hours studied (xx) and quiz score in points (yy): (1, 52), (2, 60), (3, 57), (4, 70), (5, 74), (6, 83). Find and interpret the correlation coefficient.

  1. Find the means: xˉ=216=3.5\bar{x} = \frac{21}{6} = 3.5 hours and yˉ=3966=66\bar{y} = \frac{396}{6} = 66 points.

  2. Find the sample standard deviations. The squared deviations of xx sum to 17.5, so sx=17.55≈1.8708s_x = \sqrt{\frac{17.5}{5}} \approx 1.8708 hours. The squared deviations of yy sum to 682, so sy=6825≈11.6790s_y = \sqrt{\frac{682}{5}} \approx 11.6790 points.

  3. Standardize each value. The z-scores for xx are −1.3363-1.3363, −0.8018-0.8018, −0.2673-0.2673, 0.26730.2673, 0.80180.8018, 1.33631.3363. The z-scores for yy are −1.1987-1.1987, −0.5137-0.5137, −0.7706-0.7706, 0.34250.3425, 0.68500.6850, 1.45561.4556.

  4. Multiply the paired z-scores: 1.60191.6019, 0.41190.4119, 0.20600.2060, 0.09150.0915, 0.54920.5492, 1.94511.9451. Every product is positive because each student is on the same side of the mean in both variables.

  5. Add the products: about 4.80564.8056.

  6. Divide by n−1=5n - 1 = 5: r=4.80565≈0.961r = \frac{4.8056}{5} \approx 0.961.

  7. Check with the equivalent form r=∑(xi−xˉ)(yi−yˉ)(n−1)sxsy=1055(1.8708)(11.6790)≈0.961r = \frac{\sum (x_i - \bar{x})(y_i - \bar{y})}{(n-1) s_x s_y} = \frac{105}{5(1.8708)(11.6790)} \approx 0.961.

r≈0.961r \approx 0.961. There is a strong, positive, linear association between hours studied and quiz score for these six students. This does not show that studying more causes higher scores.

Frequently asked questions

What is a good correlation coefficient?

There is no single good value. How large rr needs to be depends on the field and the purpose. In general, values closer to 11 or −1-1 indicate a stronger linear association, and the scatterplot should always be checked before trusting the number.

What does a correlation coefficient of 0 mean?

It means there is no linear association between the two variables. It does not mean there is no relationship. Points that follow a symmetric curve, such as a U shape, can have rr equal to 0 while yy is closely related to xx.

Can the correlation coefficient be greater than 1?

No. rr is always between −1-1 and 11 inclusive. If a calculation gives a value outside that range, there is an arithmetic error, often from using nn in one place and n−1n - 1 in another.

Does a negative correlation mean a weak relationship?

No. The sign gives direction and the distance from 0 gives strength. r=−0.9r = -0.9 is a stronger linear association than r=0.5r = 0.5. See positive vs negative correlation.