Course contentsShow
← Mathematics
Lesson 2950 of 3,12158. Joint and Limit BehaviorFree lesson

Covariance and Independence

Understand that independent random variables have zero covariance, but zero covariance doesn't imply independence.

Covariance and Independence

What You'll Learn

You'll understand the asymmetric relationship between independence and covariance: independent random variables always have zero covariance, but zero covariance doesn't guarantee independence.

The Core Idea

When two random variables X and Y are independent, they don't influence each other at all. Remember from earlier lessons that independence means knowing X tells you nothing about Y, and their joint distribution factors into the product of marginals.

It turns out that independence always implies zero covariance. This makes intuitive sense: covariance measures how two variables move together linearly. If X and Y are truly independent, there's no relationship to capture, so Cov(X,Y) = 0.

Here's the key subtlety: the reverse isn't true. You can have two random variables with zero covariance that are still dependent!

Why Zero Covariance Doesn't Imply Independence

Covariance only captures linear relationships. Two variables might have a strong non-linear relationship yet still have zero covariance.

Analogy: Imagine X is uniformly distributed from -1 to 1, and Y = X². These are clearly dependent—if you know X, you know Y exactly. But if you calculate the covariance, it equals zero! Why? Because Y depends on X symmetrically: positive and negative X values contribute equally but in opposite directions to the covariance calculation, canceling out.

Zero covariance means "no linear association," not "no association at all."

The Bottom Line

  • Independence → Zero Covariance (always true)
  • Zero Covariance → Independence (NOT always true)

Independence is a stronger condition that rules out all forms of dependence. Zero covariance only rules out linear dependence.

Key Takeaway: Independent random variables always have zero covariance, but zero covariance only guarantees the absence of linear relationship—non-linear dependencies can still exist.