[RsR] Comparing OLS and robust regression

[Andreas Ruckstuhl]

Dear Michael

I recommend to compare the graphical residual analysis of both fits. 
This is done best when lmrob() of the package robustbase is used because 
then you can proceed as follows

library(robustbase)
D.rlm <- lmrob(y ~ x1 + x2, data=D, setting='KS2011')
par(mfrow=c(2,3))
plot(D.rlm)


If the robust fit does not show any outliers or any other strange 
things, the classical and the robust fit should agree. If there are 
outliers you cannot trust the output of the classical fit (i.e., 
estimated coefficients, standard errors and test results). In this case 
you must use the results of the robust fit.


Best regards
Andreas


Am 22.05.2015 um 23:24 schrieb michael westphal via R-SIG-Robust:
> Hello:
> I am using R 3.0.2.
> I have built robust regression models with rlm, because regression diagnostic tests (Q-Q plots) indicate there are large outliers. How do I best compare the goodness of fit between robust regression and OLS?  Does it make sense to only compare residual standard error?  If not, what do you suggest? Thanks.
> Michael
>
> 	[[alternative HTML version deleted]]
>
> _______________________________________________
> R-SIG-Robust using r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-robust

-- 
----------------------------------------------------------------------

Prof. Dr. Andreas Ruckstuhl
ZHAW Zürcher Hochschule für Angewandte Wissenschaften
IDP Institut für Datenanalyse und Prozessdesign
Rosenstrasse 3                    Tel.  : +41 (0)58 934 78 12
Postfach                          Fax   : +41 (0)58 935 78 12
CH-8401 Winterthur                e-Mail: Andreas.Ruckstuhl using zhaw.ch
                                   WWW   : http://www.idp.zhaw.ch