In Econometrics Tutoring and Statistics Tutoring, heteroskedasticity is one of the most important violations of the Gauss–Markov assumptions. While it does not bias OLS coefficients, it does bias the standard errors, which means t‑tests, p‑values, and confidence intervals become unreliable.
This page explains what heteroskedasticity is, why it biases standard errors, and how to detect and correct it using robust methods.
What Is Heteroskedasticity?
Formally, heteroskedasticity means:
\[ Var(u_i \mid X_i) \neq \sigma^2 \]
When this assumption fails, OLS is still unbiased, but the estimated standard errors are wrong.
Why Heteroskedasticity Biases Standard Errors
1. OLS formulas assume constant variance
The standard error formulas used in OLS rely on the assumption of homoskedasticity. When variance changes across observations, these formulas no longer reflect the true sampling variability.
2. High‑variance observations distort the variance estimate
Observations with larger error variance contribute disproportionately to the residuals, inflating or deflating the estimated standard errors.
3. The OLS variance estimator becomes inconsistent
Under heteroskedasticity:
\[ \widehat{Var}(\hat{\beta}) \neq Var(\hat{\beta}) \]
This means the estimated standard errors do not converge to the true ones.
4. t‑tests and p‑values become invalid
If standard errors are wrong, hypothesis tests are wrong. You may incorrectly reject or fail to reject hypotheses.
5. Confidence intervals become too wide or too narrow
Biased standard errors lead to incorrect interval widths, undermining inference.
How Heteroskedasticity Affects Standard Errors (Step by Step)
Step 1: Write the OLS variance formula
Under homoskedasticity:
\[ Var(\hat{\beta}) = \sigma^2 (X’X)^{-1} \]
OLS uses this formula to compute standard errors.
Step 2: Replace constant variance with heteroskedastic variance
Under heteroskedasticity:
\[ Var(u_i \mid X_i) = \sigma_i^2 \]
The true variance becomes:
\[ Var(\hat{\beta}) = (X’X)^{-1} X’ \Omega X (X’X)^{-1} \]
where \( \Omega \) is a diagonal matrix of unequal variances.
Step 3: OLS still uses the wrong formula
OLS continues to use the homoskedastic formula, which no longer matches the true variance.
Step 4: Standard errors become biased
The estimated standard errors may be:
- too small → inflated t‑statistics → false positives
- too large → deflated t‑statistics → false negatives
Step 5: Use robust standard errors to fix the problem
Heteroskedasticity‑robust (White) standard errors estimate:
\[ \widehat{Var}_{robust}(\hat{\beta}) = (X’X)^{-1} X’ \widehat{\Omega} X (X’X)^{-1} \]
This produces consistent standard errors even when heteroskedasticity is present.
Numerical Example
Suppose the true model is:
\[ Y = \beta_0 + \beta_1 X + u \]
But the variance of the error term increases with X:
\[ Var(u_i \mid X_i) = 4X_i^2 \]
OLS will treat all observations as equally noisy, even though high‑X observations have much larger variance.
This leads to:
- underestimated standard errors for large‑X observations
- overconfident t‑tests
- incorrect p‑values
Using robust standard errors corrects the problem.
Common Mistakes
- Thinking heteroskedasticity biases OLS coefficients (it does not).
- Using regular standard errors when heteroskedasticity is present.
- Assuming residual plots always reveal heteroskedasticity.
- Ignoring heteroskedasticity in cross‑sectional data.
- Believing robust standard errors “fix” all problems (they do not fix bias from endogeneity).
Why This Matters
Understanding heteroskedasticity helps you:
- perform valid hypothesis testing
- avoid misleading p‑values
- construct correct confidence intervals
- diagnose model misspecification
- use robust standard errors appropriately
It is essential for reliable econometric inference.
Related Topics
This idea connects directly to:
Speak Directly to a Tutor — Send Your Message Below
No call centers. No delays. Your message goes straight to the tutor.
- Call/Text: 510-398-0006
- Email: tutor@californiagraduatetutor.com
- WhatsApp: Send Files