[R] software comparison

[Owe Jessen]

Ben Bolker schrieb:
> Philippe Grosjean <phgrosjean <at> sciviews.org> writes:
>
>   
>> Hello,
>>
>> For a paper published in 2007, and submitted in April 2005, this is 
>> still surprising. If I my calculation is correct, in 2004, they would 
>> have used R 2.2.x, or something,... not 1.9.1?
>>
>> Anyway, does someone know if there is a chance 2.5.0 provides some 
>> improvements in some of the "difficult cases" for R 1.9.1 (for instance, 
>> improvement of the algorithms for calculating autocorrelation, ANOVA, 
>> linear regression or non linear regression)?
>>
>>     
>
>   I did a bit of testing (with 2.6.0 unstable).
> Two of the tests that R 1.9.1 did worst on were
> ANOVAs of NIST data sets SiRstv and AgWt 
> (LREs listed in the paper for these tests are
>  5.7 and 6.7 respectively, compared
> with SAS 9.1's 12.7 and 8.8, Splus 6.2's 13.4
> and 9.7).
>
>   The code below produces LREs of
> 13.3 for SiRstv and 9.18 for AgWt -- close to Splus
> 6.2, better than SAS.
>
>   If anyone has a whole zoo of R versions lying around
> they could go back and try this code on them ...
>
>   Ben Bolker
>
>
>   
The results seem to be rather stable, I get identical values on R 2.4.1
and 2.2.1

HTH

Owe Jessen