Skip to content
Prev 15286 / 63421 Next

(PR#7707) Re: [Rd] vcov on result of rlm() yields "-- please report!"

This results from calling vcov() on an object for which it does not have a 
method.  That is allowed, but that you get an warning is not a bug: it 
should point up to you that the results are not reliable.  However, there 
is an error in the check!  (So the bug is not what you were asked to 
report: it seems strange to me that it asks for a report on something 
that is perfectly reasonable.)

I do not remember why vcov.lm calls summary.lm explicitly (I suspect it is 
to cope with aov objects), but that stops inheritance working correctly. 
Adding

vcov.rlm <- function (object, ...)
{
     so <- summary(object, corr = FALSE)
     so$stddev^2 * so$cov.unscaled
}

should circumvent this and also give the results you should expect.

I believe the results you get under S-PLUS are equally unreliable, for the 
same reason.  Nowhere that I can see is it stated that vcov works for rlm 
fits (do check its help page), and a leap of faith underlies this.


Finally, it is courtesy to acknowledge when you are using contributed 
software (I hope you _do_ give credit in your reports), and both the 
posting guide and FAQ do ask you to report problems with contributed 
packages to the maintainer.  (You seem unaware that rlm is part of the VR 
bundle.)  E.g.:

    Bug reports on contributed packages should be sent first to the package
    maintainer, and only submitted to the R-bugs repository by package
    maintainers, mentioning the package in the subject line.
On Fri, 25 Feb 2005 frank@trdlnk.com wrote: