Correlation — How Two Variables Move Together
Measuring the strength of a linear relationship
This concept builds on Linear Regression — The Line of Best Fit. Consider exploring it first for the best experience.
Two Variables, One Relationship
Correlation measures how two variables move together. Does taller height mean heavier weight? Do more hours studied lead to higher scores? A scatter plot reveals the pattern: each dot is one case, placed across by its X value and up by its Y value. Watch the 60 dots appear one at a time — even before any number is shown, you can see whether the cloud leans up, leans down or has no lean at all.
The Pearson Coefficient r
r ranges from −1 to +1. r=+1 is perfect positive correlation (both go up together). r=−1 is perfect negative. r=0 means no linear relationship at all. The line drawn through the cloud is the best-fit straight line, and the label beside it shows the current r. Under the hood, r multiplies each point's distance from the mean of X by its distance from the mean of Y, adds these products and scales the total so it cannot leave the −1 to +1 range.
Positive Correlation
When r > 0, high values of X tend to pair with high values of Y. The scatter plot tilts upward. Watch the points arrange themselves into a rising pattern. Points in the upper-right and lower-left of the centre add positive products to r; points in the other two corners subtract. The thinner and more tightly the cloud hugs its line, the closer r gets to +1.
Negative Correlation
When r < 0, high X pairs with low Y. The plot tilts downward. Classic examples: temperature vs. heating bills, study time vs. errors, speed vs. fuel efficiency. Notice that the sign of r only tells you the direction of the tilt; the size of r tells you how tightly the dots cluster around the line. A gentle downward slope with tightly packed points can have a larger |r| than a steep but scattered one.
No Correlation
When r ≈ 0, there is no straight-line trend: the cloud of points is round, not elongated. Drag the slider to see the shape transition, and watch the "R² (Explained)" card fall towards 0%. But r ≈ 0 does not mean "no relationship" — choose "Curved (U-shape)" from the "Data pattern" menu: Y follows X closely along a curve, yet r stays near 0, because the falling left half and the rising right half cancel out.
Correlation ≠ Causation
High r only means the variables move together — not that one causes the other. Ice cream sales and drowning rates are correlated (both peak in summer). Always ask: what's the third variable? Choose "Two hidden groups" from the "Data pattern" menu: each group slopes downward, yet the pooled r is positive. To show that X really changes Y you need a controlled experiment, where X is set on purpose and everything else is held fixed.
One Point Can Swing r
Correlation is sensitive to outliers. Try dragging a single point far from the cloud — watch r shift dramatically even though 59 other points barely moved. Click "New Sample" to redraw 60 fresh points at the same target r and see how much the actual sample r wobbles around it — real data rarely matches its population r exactly. Then click "New Challenge", set "Your guess" and press "Check Guess" to test how well you read r by eye.
How It Works — Step by Step
-
1
Two Variables, One Relationship
Correlation measures how two variables move together. Does taller height mean heavier weight? Do more hours studied lead to higher scores? A scatter plot reveals the pattern: each dot is one case, placed across by its X value and up by its Y value. Watch the 60 dots appear one at a time — even before any number is shown, you can see whether the cloud leans up, leans down or has no lean at all.
-
2
The Pearson Coefficient r
r ranges from −1 to +1. r=+1 is perfect positive correlation (both go up together). r=−1 is perfect negative. r=0 means no linear relationship at all. The line drawn through the cloud is the best-fit straight line, and the label beside it shows the current r. Under the hood, r multiplies each point's distance from the mean of X by its distance from the mean of Y, adds these products and scales the total so it cannot leave the −1 to +1 range.
-
3
Positive Correlation
When r > 0, high values of X tend to pair with high values of Y. The scatter plot tilts upward. Watch the points arrange themselves into a rising pattern. Points in the upper-right and lower-left of the centre add positive products to r; points in the other two corners subtract. The thinner and more tightly the cloud hugs its line, the closer r gets to +1.
-
4
Negative Correlation
When r < 0, high X pairs with low Y. The plot tilts downward. Classic examples: temperature vs. heating bills, study time vs. errors, speed vs. fuel efficiency. Notice that the sign of r only tells you the direction of the tilt; the size of r tells you how tightly the dots cluster around the line. A gentle downward slope with tightly packed points can have a larger |r| than a steep but scattered one.
-
5
No Correlation
When r ≈ 0, there is no straight-line trend: the cloud of points is round, not elongated. Drag the slider to see the shape transition, and watch the "R² (Explained)" card fall towards 0%. But r ≈ 0 does not mean "no relationship" — choose "Curved (U-shape)" from the "Data pattern" menu: Y follows X closely along a curve, yet r stays near 0, because the falling left half and the rising right half cancel out.
-
6
Correlation ≠ Causation
High r only means the variables move together — not that one causes the other. Ice cream sales and drowning rates are correlated (both peak in summer). Always ask: what's the third variable? Choose "Two hidden groups" from the "Data pattern" menu: each group slopes downward, yet the pooled r is positive. To show that X really changes Y you need a controlled experiment, where X is set on purpose and everything else is held fixed.
-
7
One Point Can Swing r
Correlation is sensitive to outliers. Try dragging a single point far from the cloud — watch r shift dramatically even though 59 other points barely moved. Click "New Sample" to redraw 60 fresh points at the same target r and see how much the actual sample r wobbles around it — real data rarely matches its population r exactly. Then click "New Challenge", set "Your guess" and press "Check Guess" to test how well you read r by eye.
Key Takeaways
r Is Bounded by −1 and +1
|r| = 1 means perfect linear relationship. |r| = 0 means none. In practice, |r| > 0.7 is considered strong, 0.3–0.7 moderate, < 0.3 weak.
Shape of the Scatter Plot
Strong correlation → elongated ellipse. Weak correlation → circular cloud. Negative r → ellipse tilts downward. A curved or clustered cloud can hide a strong relationship behind a small or misleading r, so always look at the plot as well as the number.
R² Tells You Explained Variance
r² (r-squared) is the fraction of variance in Y explained by X. r=0.8 → r²=0.64 → X explains 64% of Y's variance.
Correlation ≠ Causation
Correlation is symmetric (r(X,Y) = r(Y,X)) and observational. Causation requires experimental design. Never skip this distinction.
Connected Concepts
Linear Regression
R² (R-squared) equals the square of the correlation coefficient — how well the line explains the variance.
Normal Distribution
Pearson correlation assumes both variables are approximately normally distributed.
Hypothesis Testing
You can test whether a correlation is statistically significant with a t-test on r.
Simpson's Paradox
Pick "Two hidden groups" here: the trend inside each group reverses when the groups are pooled — the same flip Simpson's Paradox shows with totals.
Z-Score
r is the average product of paired z-scores — which is why changing units never changes r.
Updated
- — Fuller step explanations of how each point pushes r up or down, plus a link to z-scores.
- — Added a "Data pattern" menu (curved and two-group data) and a "Guess r" challenge, with two new quiz questions.