Re: bounded influence regression Krasker & Welsch

From: Richard Ulrich (Rich.Ulrich_at_comcast.net)
Date: 09/29/04 

Date: Wed, 29 Sep 2004 12:14:37 -0400

 - I don't know anything about the paper or the algorithm,
or what kind of regression it is. In OLS regression, it is
a beginner's error to think that omitting the intercept term
is a trivial matter, rather than a drastic mistake.

On 27 Sep 2004 21:28:55 -0700, k2lau@netscape.net (Karen Lau) wrote:

> hi all
>
> I am trying to implement the "efficient bounded-influence regression
> estimation" algorithm in Krasker & Welsch 1982 paper. The authors
> stated that the scale estimate sigma_n is nonrobust for regression
> through the origin, does anyone have any suggestion how to work around
> that? The regression model I am working on does not have a constant
> term and when I ran the algorithm, the scale parameter just go to
> zero.
>
> thanks for any help

I suppose it could be that you have mis-programmed the algorithm,
if you did it yourself. On the other hand, that result could be
exactly what the authors had in mind when they warned about
it not being robust. (Is this in a stat-package, with
documentation?)

What do you mean by, "Work around that"? The direct
interpretation seems to be that (a) the model is seriously
mis-specified, or (b) it can't be solved by the algorithm.

Situation (a) and (b) may or may not be the same thing.
That is, the problem might make logical sense, but this
"efficient" algorithm can't deal with the form. In that case,
I guess you need a new algorithm, or an older, less
efficient algorithm. And I can't help with that. 

-- 
Rich Ulrich, wpilib@pitt.edu
http://www.pitt.edu/~wpilib/index.html

Relevant Pages

  • Re: Regression analysis -- how-to?
    ... > were it not for the fact that, the last time I asked an algorithm ... Do you want to learn about regression? ... > (linear, quadratic, logarithmic, exponential, power and inverse) and at ... > least two types of linear correlation coefficient (rank and product ...
    (comp.lang.pascal.delphi.misc)
  • Re: results too good?
    ... Artificial neural networks are not a random process. ... Use 7-fold cross-validation where regression MSE or classification ... Repeating this whole algorithm repeatedly ... increased averaging smoothen my line even more? ...
    (comp.soft-sys.matlab)
  • Re: Help on Partial Least Squares Aalgorithm
    ... > The PLS algorithm does regression on some ... > more than one target variables and give me the regression coefficent ... Now there are ways to rewrite the PLS algorithm using an eigenvector ...
    (sci.stat.math)
  • Re: Help on Partial Least Squares Aalgorithm
    ... Paige Miller wrote: ... >> Q are the kx1 regression coefficents. ... >> more than one target variables and give me the regression coefficent ... > even more calculations than the NIPALS or SIMPLS algorithm anyway. ...
    (sci.stat.math)
  • Re: Translation from recursive to iterative
    ... >>probably an efficient algorithm. ... >>reasonable number of slaves peppered through the range. ... > stack overflow. ...
    (comp.lang.c)