OLS assumptions after estimating robust standard errors

Hello everybody,

To counter problems with outliers and heteroscedasticity, I have computed an OLS with robust standard errors (using the "robust" function in Stata)

Could anybody tell me which assumptions of OLS I would still need to test for? / which of the assumptions can be relaxed when computing a "robust" model?


Hi andony:

I think when you use OLS for the regression analysis, you need to focus on the "residual plots" because the Gauss-Markov assumptions should be held for using Ordinary Least Square method. The assumptions are:

1. E(e) = 0
2. Cov(e) = sigma^2*I (I is the identity matrix)

You may need to plot the residuals against the fitted values, as well as "leverage plots" to see if you can delete any outliers and influential points. Be cautious about any deletion! Sometimes an outlier may be very informative!

You may try ridge regression if there exists multicollinearity problem. In sum, to analyze a regression problem requires comprehensive consideration.

Good luck!!