[RsR] Library for regression with t-distributed data

Southworth, Harry H@rry@Southworth @end|ng |rom @@tr@zenec@@com
Tue May 9 12:17:17 CEST 2006


Dear list,

I'm a noob. If there is already a library/package for regression with
t-distributed residuals, please forgive me and ignore me.

I have an S-PLUS library for fitting, and studying regression models under
the assumption of t-distributed data. I never got around to fully validating
it, so never made it publicly available. I sometimes use the functions and
they appear to work.

There are functions for printing, plotting and summarizing. Also, there are
functions for computing profile likelihoods. The library lets the user
decide if they want to fix the degrees of freedom parameter, or let it vary.
IWLS is used, as described by Lange, Little and Taylor.

>From memory, I implemented the option to use the expected information, but
was never convinced that I did it correctly (I don't recall why I
implemented it in the first place). Also there is no help system. Apart from
that, it's ok but needs a bit more testing out in the wild.

All of the code is S3 - there is no precompiled C or anything else. It ought
to be straightforward to port to R.

I've sat on this library for a long time and am unlikely ever to complete it
on my own. Therefore, if anyone would like to assume ownership of it and
port to R, or if anyone would like to collaborate with me on finalizing it ,
I would be very grateful.

Thanks,
Harry.

____________________________________________

Harry Southworth,
FE2 A/2 Parklands,
Alderley Park,
Cheshire, SK10 4TG, UK
Tel.: +44 1625 518327
Mobile: 07867 676073




More information about the R-SIG-Robust mailing list