01 / 21

Understanding
Correlation

Explore how the correlation coefficient \( r \) measures the strength and direction of linear relationships between two quantitative variables.

AP Statistics

What Is Correlation?

Correlation describes the strength and direction of a linear relationship between two quantitative variables. We measure it with the correlation coefficient \( r \), a single number between \( -1 \) and \( 1 \).

Direction
Positive \( r \): both variables increase together. Negative \( r \): one increases as the other decreases.
Strength
Values of \( r \) near \( \pm 1 \) indicate a strong linear association. Values near \( 0 \) indicate a weak one.

Visualizing Correlation in Scatterplots

Strong Positive
\( r \approx 0.95 \)
Weak Positive
\( r \approx 0.35 \)
No Correlation
\( r \approx 0.02 \)
Strong Negative
\( r \approx -0.92 \)

Why Correlation Matters

Correlation is foundational to regression, prediction, and inference in statistics. Before fitting a line to data, we check \( r \) to ensure a linear model is appropriate.

Prediction
A strong \( r \) justifies using the least-squares regression line to predict \( y \) from \( x \).
Screening
Researchers use \( r \) to identify which pairs of variables are worth investigating further.
Communication
A single number \( r \) summarizes a relationship concisely for reports and publications.

The Formula for \( r \)

1
Standardize each value: convert every \( x_i \) and \( y_i \) into z-scores using their respective means and standard deviations.
2
Multiply paired z-scores: for each observation, compute \( z_{x_i} \cdot z_{y_i} \).
3
Average the products: divide the sum by \( n - 1 \) to get \( r \).

Key Properties of \( r \)

1
Bounded: \( -1 \leq r \leq 1 \). Values of \( \pm 1 \) mean all points lie exactly on a line.
2
Unitless: because we standardize, \( r \) has no units and does not depend on the scale of measurement.
3
Symmetric: the correlation of \( x \) with \( y \) equals the correlation of \( y \) with \( x \).
4
Not resistant: \( r \) is sensitive to outliers. A single extreme point can dramatically change \( r \).
5
Linear only: \( r \) measures only linear association. A perfect curve can yield \( r \approx 0 \).

\( r \) vs \( r^2 \): Key Distinction

\( r \) (Correlation) \( r^2 \) (Coefficient of Determination)
Range \( -1 \) to \( 1 \) \( 0 \) to \( 1 \)
Sign Indicates direction (+ or −) Always non-negative
Interpretation Strength and direction of linear relationship Proportion of variance in \( y \) explained by \( x \)
Example \( r = -0.80 \) \( r^2 = 0.64 \): 64% of variation explained

Interpreting the Strength of \( r \)

\( -1 \)
\( -0.5 \)
\( 0 \)
\( 0.5 \)
\( 1 \)
Strong − Moderate − Weak Weak Moderate + Strong +
Rough guidelines: \( |r| > 0.8 \) strong, 0.5–0.8 moderate, \( |r| < 0.5 \) weak

How to Calculate \( r \)

1
Make a scatterplot to check that the relationship looks linear and has no extreme outliers.
2
Find the means \( \bar{x} \) and \( \bar{y} \), and the standard deviations \( s_x \) and \( s_y \).
3
Standardize: compute \( z_{x_i} = \frac{x_i - \bar{x}}{s_x} \) and \( z_{y_i} = \frac{y_i - \bar{y}}{s_y} \) for each data point.
4
Multiply each pair: \( z_{x_i} \cdot z_{y_i} \).
5
Sum and divide by \( n - 1 \): \( r = \frac{1}{n-1}\sum z_{x_i} z_{y_i} \).

Example 1: Study Hours vs Test Score

0 1 2 3 4 5 Hours Studied 50 60 70 80 90 100 Score
\( r = 0.96 \)

Example 2: Negative Correlation

A researcher measures the age of a car (years) and its resale value (thousands of dollars) for 8 cars.

0 2 4 6 8 10 Age (years) 0 10 20 30 Value ($K)
\( r = -0.99 \)

Example 3: When \( r \) Is Misleading

x y
\( r \approx 0.00 \)
Perfect Parabola: \( r \approx 0 \)

Correlation Does Not Imply Causation

A strong correlation between two variables does not mean one causes the other. There are several alternative explanations:

Lurking Variable
A hidden third variable drives changes in both \( x \) and \( y \), creating a spurious correlation.
Confounding
The effect of one variable is mixed with the effect of another, making it impossible to separate causes.
Coincidence
With enough variable pairs, some will correlate by pure chance: ice cream sales correlate with drowning rates.

Outliers and Influential Points

Outlier
\( r = 0.99 \) (without)
\( r = 0.41 \) (with)

Common Mistakes with Correlation

✕ Claiming Causation
Saying "x causes y" from correlation alone. Only randomized experiments establish causation.
✕ Ignoring Nonlinearity
Reporting \( r \) for data that follows a curve. Always plot data first to verify linearity.
✕ Forgetting Context
Stating "\( r = 0.85 \)" without naming the variables. AP requires: "There is a strong positive linear relationship between [x] and [y]."
✕ Misreading \( r^2 \)
Confusing \( r = 0.7 \) with 70% explained. Actually \( r^2 = 0.49 \), so only 49% of variance is explained.

Summary: Correlation at a Glance

Concept Key Point
Symbol \( r \) (sample), \( \rho \) (population)
Range \( -1 \leq r \leq 1 \)
Measures Strength and direction of linear association
Units None: \( r \) is unitless
\( r^2 \) Proportion of variance in \( y \) explained by \( x \)
Sensitivity Not resistant to outliers
Limitation Does not imply causation

Question 1 of 5

A scatterplot of weekly exercise hours vs resting heart rate shows a strong downward trend. Which value of \( r \) is most plausible?

Question 2 of 5

If the correlation between height (in inches) and weight (in pounds) is \( r = 0.72 \), what is \( r \) when height is measured in centimeters instead?

Question 3 of 5

A dataset has \( r = 0.60 \). What is the correct interpretation of \( r^2 \)?

Question 4 of 5

Data on ice cream sales and sunburn rates show \( r = 0.92 \). Which conclusion is most appropriate?

Question 5 of 5

A researcher removes a clear outlier from a dataset and \( r \) changes from \( 0.40 \) to \( 0.91 \). What does this demonstrate?