Course contentsShow
← Data Science
Lesson 553 of 2,14514. Simple Linear RegressionFree lesson

Exogeneity: X Must Be Independent of Errors

Understanding why the predictor cannot be correlated with the error term for valid inference.

Exogeneity: X Must Be Independent of Errors

You'll learn: Why your predictor variable must not be influenced by — or share hidden causes with — the error term in your regression model.

The Core Idea

Exogeneity means that your predictor variable X is determined outside the model and is completely independent of the error term ε. In other words, knowing the value of X gives you no information about the error, and vice versa.

Mathematically, this is expressed as: Cov(X, ε) = 0

This is closely related to the "zero conditional mean" assumption you just learned, but goes further: X cannot be correlated with errors at all.

Why It Matters

When X and the errors are correlated, your slope estimate β₁ becomes biased. The model attributes variation to X that actually belongs to omitted factors hiding in the error term.

Common violations:

  • Reverse causality: Y affects X, not just X affecting Y
  • Omitted variable bias: A third variable influences both X and Y, creating a spurious relationship
  • Measurement error in X: Recording mistakes correlate with unpredictable noise
  • Simultaneity: X and Y determine each other simultaneously

An Analogy

Imagine measuring how study hours (X) affect test scores (Y). If students who study more also tend to have private tutors (hidden in the error term), then X is correlated with ε. Your regression will wrongly credit all the score improvement to study hours alone, overestimating its effect.

Key Takeaway: Exogeneity requires that your predictor is "clean" — truly independent from all the unmeasured factors in your error term. Violations lead to biased, unreliable coefficient estimates.