MATLAB Central Newsreader - RobustFit goodness of fit

RobustFit goodness of fit

rivoli2@gmail.com — Mon, 15 Dec 2008 12:11:27 -0500

Dear Listers,
I'd like to know how to 'understand' the goodness of fit of a Robust
regression by the stats result (r,s,robust_s, etc)

Any suggestion would be greatly appreciated.

Cristiano.

Re: RobustFit goodness of fit

Tom Lane — Mon, 15 Dec 2008 18:39:37 -0500

> I'd like to know how to 'understand' the goodness of fit of a Robust
> regression by the stats result (r,s,robust_s, etc)

Cristiano, if you type "edit robustfit" and look below the regular help
text, you will find some additional comments listing references. The paper
by DuMouchel and O'Brien talks about this issue.

-- Tom

Re: RobustFit goodness of fit

Stephen — Sat, 04 Jul 2009 04:11:01 -0400

"Tom Lane" <tlane@mathworks.com> wrote in message <gi6899$mee$1@fred.mathworks.com>...
> > I'd like to know how to 'understand' the goodness of fit of a Robust
> > regression by the stats result (r,s,robust_s, etc)
>
> Cristiano, if you type "edit robustfit" and look below the regular help
> text, you will find some additional comments listing references. The paper
> by DuMouchel and O'Brien talks about this issue.
>
> -- Tom
>

The robust fitting and curvefit functions are incredibly useful, but...it would be nice if the help for the functions went a little farther in explaining what was being calculated. For example, it is not clear what r^2 means for robust fitting since it must be different from what is done for OLS and presumably incorporates the weighting used in the fit. Since you have coded some particular equations in the m-file, why not write them down? Also - it would be nice if robustfit.m included the r^2 that is given in curvefit.
Stephen
ps: unfortunately the DuMouchel and O'Brien paper seems to be the rare case of something not available online.

Re: RobustFit goodness of fit

Mike — Wed, 24 Nov 2010 16:22:07 -0500

I have similar questions about robustfit. In particular, how are the following calculated?

ols_s — Sigma estimate (RMSE) from ordinary least squares

robust_s — Robust estimate of sigma

s — Final estimate of sigma, the larger of robust_s and a weighted average of ols_s and robust_s

I suppose s is an attempt to define a consistent metric for goodness-of-fit? Is this the actual criterion used for the termination rule?

I want to do a fitting where robustfit is in the inner loop and a parameter that transforms the explanatory variables is adjusted in the outer loop, but dont know what gof statistic to use in the outer minimization.

Any help greatly appreciated.

"Stephen " <removethis.monismith@stanford.edu> wrote in message <h2mkol$83o$1@fred.mathworks.com>...
> "Tom Lane" <tlane@mathworks.com> wrote in message <gi6899$mee$1@fred.mathworks.com>...
> > > I'd like to know how to 'understand' the goodness of fit of a Robust
> > > regression by the stats result (r,s,robust_s, etc)
> >
> > Cristiano, if you type "edit robustfit" and look below the regular help
> > text, you will find some additional comments listing references. The paper
> > by DuMouchel and O'Brien talks about this issue.
> >
> > -- Tom
> >
>
> The robust fitting and curvefit functions are incredibly useful, but...it would be nice if the help for the functions went a little farther in explaining what was being calculated. For example, it is not clear what r^2 means for robust fitting since it must be different from what is done for OLS and presumably incorporates the weighting used in the fit. Since you have coded some particular equations in the m-file, why not write them down? Also - it would be nice if robustfit.m included the r^2 that is given in curvefit.
> Stephen
> ps: unfortunately the DuMouchel and O'Brien paper seems to be the rare case of something not available online.