Theorem
Points on one line have or
Statistics • Correlation and Regression
Statement
If the observations , with the not all equal, all satisfy with , then the correlation coefficient of and is when and when . It can never be or any value strictly between and .
How to use it
- 1 is after a change of origin and scale. Such a change keeps the size of the correlation and only flips its sign when the scale is negative.
- 2The correlation of with itself is . So the correlation of with is for a rising line and for a falling one.
- 3 would need no linear trend at all; a line with non-zero slope is the strongest trend there is.
Worked example
The points , , , are given. Find .
Step-by-step
- 1They satisfy
Why: Check the slope between each pair: always .
- 2Slope
Why: A falling line.
- 3
Answer
Watch out
A horizontal line () makes all equal, so and is not defined at all; it is not .
Builds on
Also known as
perfect correlation, correlation of points on a straight line, r = plus or minus 1