> since the OLS and robust regressions have the same number of DFs, looking
> at the residual standard error is insightful.
Sadly not. The residual scale in a robust model is only partly indicative of
goodness of fit; robust models intentionally downweight outliers. Much of the
difference in scale can be due to downweighting, rather than change in model,
especially where outliers are roughly symmetricaly distributed. And the degrees
of freedom are not, strictly, the same. You have the same numbers of
observations, but once you throw in different weights, it's debatable whether
the effective df are really equal to the classical df. In any case degrees of
freedom mostly matters as a distribution parameter - if you could trust the
distribution to be normal, chi-squared etc you would not need robust statistics.
What you can do, to an extent, is use something like lmRob in the robustbase
package to test your fixed effects; comparing the different inferences will
tell you something about which effects in OLS are simply artefacts caused by
outliers. lmRob uses comparatively recent developments in wald-type inference
tests to put the tests on a firmer footing.
S Ellison
*******************************************************************
This email and any attachments are confidential. Any use...{{dropped:8}}
______________________________________________
R-help@r-project.org mailing list -- To UNSUBSCRIBE and more, see
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.
