[RsR] R function

Ricardo Maronna rm@ronn@ @end|ng |rom m@||@ret|n@@@r
Tue Dec 6 01:02:22 CET 2005

> Ricardo, can you be more specific about what you don't like in the
> implementation of OGK in S-Plus?

Hi Kjell,
    The "pure" OGK estimate works regardless of the collinearity of the 
    The resulting estimate needs some improvement step. We thought of the 
simple "one-step hard rejection". It was indicated in the paper that no 
explicit matrix inversion is required for this step, so that it can be made 
to work even for collinear data.
    I have tried the S-Plus version with data that were definitely not 
collinear, and I got an error message of the sort "matrix not invertible". I 
couldn't figure out why.

    By the way, on later reflexion I have thought that a better improvement 
would be to use the "pure" OGK as the starting point for the iterations of 
an S-estimate, which would be much cheaper than subsampling.

More information about the R-SIG-Robust mailing list