Properties of Correlation
You'll learn the mathematical properties that make correlation a standardized, interpretable measure of linear association between random variables.
The Range and Interpretation
The correlation coefficient ρ (or r for sample data) always lies between -1 and 1, inclusive. This bounded range makes correlation superior to covariance for interpretation:
- ρ = 1: Perfect positive linear relationship (as X increases, Y increases proportionally)
- ρ = -1: Perfect negative linear relationship (as X increases, Y decreases proportionally)
- ρ = 0: No linear relationship (though nonlinear relationships may exist)
- Values between: Indicate the strength and direction of linear association
Think of correlation like a normalized score on a test—it's always between 0% and 100% (or -100% to +100%), making comparisons meaningful regardless of the original scale.
Invariance Under Linear Transformations
Here's the remarkable property: correlation is invariant under linear transformations with non-zero slope. If you transform X to aX + b and Y to cY + d (where a, c ≠ 0), the correlation remains unchanged in magnitude:
- If a and c have the same sign, ρ stays the same
- If they have opposite signs, ρ changes sign
This means measuring temperature in Celsius or Fahrenheit, or income in dollars versus thousands of dollars, won't change the correlation. The relationship's strength and direction are preserved.
Key Takeaway: Correlation is bounded between -1 and 1, provides standardized measurement of linear association, and remains invariant under linear rescaling—making it a universal measure of linear dependence that transcends units of measurement.