Homoscedasticity: Constant Variance of Residuals
What you'll learn: You'll understand why the spread of your prediction errors should stay consistent across all values of your independent variable.
What Is Homoscedasticity?
Homoscedasticity means that the variance of your residuals (prediction errors) remains constant across all fitted values. In simpler terms: whether you're predicting small or large values, the "scatter" of your errors around the regression line should look roughly the same.
Think of it like a highway with guardrails. Homoscedasticity means the guardrails stay the same distance apart along the entire road. If the guardrails suddenly widen at one end, you'd have heteroscedasticity (non-constant variance) — a violation of this assumption.
Why Does This Matter?
Linear regression assumes that the model's uncertainty is consistent everywhere. When residuals have constant variance:
- Your predictions are equally reliable across the range of X values
- Standard errors and confidence intervals are valid
- Hypothesis tests (like testing if the slope equals zero) produce trustworthy results
When variance changes — say, errors are tiny for small X but huge for large X — the regression gives false confidence in some predictions and underestimates uncertainty in others.
Spotting Homoscedasticity
You check this visually by plotting residuals versus fitted values. You're looking for:
- ✅ A random cloud with consistent vertical spread
- ❌ A funnel shape (widening or narrowing)
- ❌ Any systematic pattern in the spread
If the vertical scatter stays roughly uniform as you move left to right, you've got homoscedasticity.
Key Takeaway: Homoscedasticity ensures your residuals have constant variance across all predictions, making your model's uncertainty estimates reliable and your statistical tests valid.