[R] software comparison

Ben Bolker bolker at zoo.ufl.edu
Mon Apr 16 20:18:13 CEST 2007


Thomas Lumley <tlumley <at> u.washington.edu> writes:

> 
> On Mon, 16 Apr 2007, Philippe Grosjean wrote:
> 
> > Hello,
> >

  [snip]

> No -- there is a new x.y.0 twice per year. Checking the r-announce 
> archives shows that 2.0.0 came out in October 2004 and 1.9.1 in June 2004.
> 
> Describing 1.9.1 as "the latest version" was inaccurate, but it may well 
> have been a more recent version than for most of the  other packages they 
> examined.
> 
>  	-thomas
> 
> Thomas Lumley			Assoc. Professor, Biostatistics
>

  This brings up another couple of questions.
  (1) do you suppose the authors have automated their procedure sufficiently
well that it would be easy to update the study with more recent versions
of software?

  (2) at what point do differences in precision become
unimportant?  It's fair to say that LRE<3 is somewhat alarming,
but is the difference between LRE=9 and LRE=10 all that important?
Certainly this difference is overwhelmed by model uncertainty.
Or is the idea that the numerical precision we can see in these
cases is the tip of the iceberg and indicates the overall level
of trust we have in the software?

   Ben



More information about the R-help mailing list