2

I'm using R to compute robust multiple linear regression. I use the command rlm from the package MASS.

As psi function I use psi.huber or psi.bisquare.

Is there a way to get an estimator of the goodness of fit of the model? Maybe something comparable to the Adjusted R-squared, for the parametric multiple linear regression?

Moreover, am I right saying that this kind of robust regression doesn't solve the problem of non-normality of the dependent or independent variables?

Forinstance
  • 755
  • 2
  • 8
  • 29
  • 3
    you should use the more modern implementations, in the robustbase package. That one will return a more complete output. For the two other questions, what do you mean by " problem of non-normality of the dependent or independent variables"? – user603 Mar 26 '14 at 20:29
  • I was thinking to take the last question away, but since you ask. I am concerned by the fact that normality test are failing when I apply them to dependent and independent variables. Can you link a guide to how to use the robustbase package? – Forinstance Mar 26 '14 at 20:58
  • And, when I use `lmrob` (robustbase) I obtain different results, even if I get an R-sqaured, from those that I obtain with `rlm`. Is there a way to use `psi.huber` or `psi.bisquare` with `lmrob`? – Forinstance Mar 26 '14 at 23:08
  • 2
    Concerning your first comment, I direct you to this [discussion](http://stats.stackexchange.com/questions/86835/normality-assumption-in-linear-regression). The question in your second comment is now a pure programming question that should perhaps be moved to stackoverflow. Please update your question to specify what is being asked. – user603 Mar 27 '14 at 09:17

0 Answers0