[R] White's test again

[Achim Zeileis]

On Fri, 9 Nov 2007, David Kaplan wrote:

> Hi all,
>
> It seems that I can get White's (HC3) test using MASS.   The syntax I
> used for the particular problem is
>
> anova(scireg3, white.adjust="hc3")

I don't think this is true. My guess is that you are using Anova() from
package "car".

> where scireg3 is an object from the lm function.  But, the anova summary
> table is all I get.  I don't get the new estimates or standard errors
> correcting for heteroskedasticity.  Is there a way to get that information?

Look at
  vignette("sandwich", package = "sandwich")
which discusses various kinds of sandwich covariances (including HC3). You
can plug these into:
  - coeftest() from package "lmtest": partial Wald tests for each
    coefficients (as in the usual summary()),
  - waldtest() from package "lmtest": Wald tests for nested models
    (similar to anova())
  - linear.hypothesis() from package "car": Wald tests for linear
    hypotheses.

Best,
Z