8. Heteroskedasticity Flashcards
what does homoskedasticity mean?
They are, you has the same variance, given any of the explanatory variables
Why is the variance so important?
Because it gives you a measure of your precision
Why is the homoskedasticity hard to satisfy in the wage and education example?
Homoskedasticity would imply that the variability in wage about its mean is constant across all education levels
How does the estimation of the coefficient change under heteroskedasticity?
E(B^)= B still holds and is still unbiased but you can no longer derive the variance in the same waty
What does not change under heteroskedasticity for OLS?
- OLS is still unbiased and consistent under heteroskedasticity
- The interpretation of R^2 does not change
What DOES change under heteroskedasticity for OLS?
- Heteroskedasticity invalidates variance, formulas for OLS estimators
- the usual F tests and t-tests are not valid under heteroskedasticity
- OLS is no longer the best linear and biased estimator
What assumptions need to be satisfied for an estimator to be unbiased?
Linear in Parameters
Random sampling
Sample variation in the explanatory variable
Zero conditional mean
What are robust standard errors?
Another way of calculating the variance so that it still holds under heteroskedasticity. They hold under homo and heteroskedasticity
What are the limitations to using robust standard errors?
Under robust standard errors, no longer directly follows a t-distribution but this does not matter under a large sample.
What is considered a large sample?
More than 120
What is a good rule of thumb about the f-value difference between using the robust se and regular se?
If there is strong heteroskedasticity differences may be larger to be on the safe side, it is advisable to always compute, robust standard errors
How do you compute robust standard errors in stata?
, vce robust
How can you get heteroskedasticity robust f-tests in stata?
Using the command test
What is an auxiliary regression?
Auxiliary regressions are regressions that are not part of your main model
What are the issues with using a white test?
The degrees of freedom become an issue particularly with large models
