On Mon, 14 Apr 2014, ChrisR wrote:
Hi everyone,
I am using the hetglm() command from the package 'glmx' (0.1-0). It seems
that hetglm() is incompatible with the robust standard errors estimator
provided in the 'AER' package: coeftest(mymodel,vcov=vcovHC)
Any suggestions how I could obtain robust standard errors for the
heteroscedastic probit?
Use
coeftest(mymodel, vcov = sandwich)
but the usual caveat applies that it is not quite clear what this is
robust against. If the some aspect of the model likelihood is misspecified
then the estimating equations are also misspecified and the estimator
itself inconsistent. This is briefly pointed out at the end of the
discussion in vignette("sandwich-OOP", package = "sandwich").
As for the difference between vcovHC() and sandwich():
- The HC1, HC2, HC3 methods from the vcovHC() function can only be applied
if there is a linear predictor plus a useful measure of leverage etc.
Thus, it cannot be applied to general heteroskedastic probit models (or it
wouldn't be clear to me how...).
- The basic sandwich() estimator, however, can be applied easily. This
uses just the so-called outer product of gradients. In the linear
regression model it corresponds to the basic HC0 estimator.
hth,
Z
Thanks,
Chris
--
View this message in context:
http://r.789695.n4.nabble.com/hetglm-and-robust-standard-errors-tp4688737.html
Sent from the R help mailing list archive at Nabble.com.
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
