Zero Conditional Mean of Errors
What you'll learn: You'll understand why residuals must average to zero at every X value for valid regression inference.
What Does "Zero Conditional Mean" Mean?
In simple linear regression, we assume that for any given value of X, the expected value of the residuals (errors) is zero. Written mathematically: E(ε | X) = 0.
This means if you picked any specific X value—say, X = 5—and could magically observe all possible residuals at that point, their average would be zero. Not just the overall average across all data points (which least squares already guarantees), but the average at each and every X value.
Why This Assumption Matters
If residuals systematically lean positive or negative at certain X values, your model is biased. You're consistently over-predicting or under-predicting in those regions, meaning your slope or intercept estimates are wrong.
An Analogy
Imagine a basketball player aiming at the hoop. The "zero conditional mean" assumption is like saying: "On average, their shots aimed from the free-throw line should center on the basket—not consistently left or right." If they always miss left from that spot, there's a systematic bias, not just random variation.
Implications
- When satisfied: Your β₀ and β₁ estimates are unbiased—on average, they hit the true population values.
- When violated: Predictions are systematically wrong at certain X ranges; coefficient estimates are misleading.
- How to check: Look at residual plots. Residuals should scatter randomly around zero across all fitted values, with no U-shape, curve, or trend.
Key Takeaway: The zero conditional mean assumption ensures your regression line isn't systematically wrong at any X value—it's the foundation of unbiased coefficient estimates.