Does regress, robust use weighted least square or ordinary least square?

[Dennis Hansen]

Dear all,

I ran a multiple linear regression model using the regress, robust function as i wanted to account for heteroscedasticity

I am wondering if the parameters are estimated using WLS or OLS with i use this function?

Can anyone help me with clarifying this.

All the best,

[Carlo Lazzaro]

Dennis:
I would say WLS. This topic is widely covered in https://www.stata.com/bookstore/micr...metrics-stata/, Chapter 5.
That said, you can use -regress- with the -robust- option to guard against heteroskedasticity.
See also -_robust- entry in Stata .pdf manual.

[Phil Bromiley]

This is also a case where it is easy to find out - run reg with and without the robust errors and see if the parameters on the variables change. 'The empirical answer is that robust changes the standard error but not the b's so I don't think it is doing WLS when you pick robust. The same happens if you use clustered robust standard errors.

If you want WLS, you need to use weights.