R help - Nov 2007 - R validation

Dear R-Users,

A message to continue the discussions we had in March and June. I have
read the documents written since then
(http://www.r-project.org/doc/R-FDA.pdf ;
http://user2007.org/program/presentations/harrell.pdf ;
http://user2007.org/program/presentations/rossini.pdf ;
http://user2007.org/program/presentations/soukup.pdf) which are very
interesting and useful.

I work in a CRO and we strongly believe that R could be our standard
statistical tool for all our analyses. I agree that it is more relevant
(and easier) to validate an analysis than the whole software but anyway,
if we want to systematically use R, we must convince our top management
first and then our clients that R is reliable. Most of them do not know
the R environment and it is more that likely that we will be asked for any
validation documentation.

I am sure that the Core team has thoroughly tested ?Base R? functions and
the recommended packages. Is it possible to have access to the scripts or
any documents that would attest that R gives reliable/validated results ?
All the discussions, documents mentioned above are a big step for the use
of R in the pharma/biotech environment but I also think that validation
documents or scripts are indispensable to convince our top management, our
clients and regulatory bodies.

If validation documentation or scripts do not exist or are not available,
could anyone tell me in concrete terms which are the minimum requirements
to validate R functions or packages ?

Thanks !

Delphine Fontaine

Delphine Fontaine
Statistician
Data & Statistics Department - Genexion SA
------------------------------------------------------------
29, Quai du Mont-Blanc
Geneva, CH-1201
Switzerland
------------------------------------------------------------
Office: +41 22 704 32 44
Fax:    +41 22 704 32 42
Email: Delphine.Fontaine at genexion.com

For the survival package, you might want to reference appendix E of Therneau 
and Grambsh, Modeling Survival Data, Springer-Verlag, 2000.  This contains a set
of small example data sets for which results have been worked out in closed 
form.  S-plus and R pass all the tests: they are a portion of the test suite of 
the package.  (SAS does not pass, BTW.  It doesn't compute dfbeta residuals
for
a Cox model quite right under the Efron approximation.  The error is small and 
almost certainly unimportant in practical data, but I like to keep it in hand as
a counterexample to believers of SAS infallability).
 
 	Terry Therneau

In regard to validation of R, there are test directories with test 
code. See for example:
https://svn.r-project.org/R/trunk/tests/
and
https://svn.r-project.org/R/trunk/src/library/stats/tests/

David Scott


_________________________________________________________________
David Scott	Department of Statistics, Tamaki Campus
 		The University of Auckland, PB 92019
 		Auckland 1142,    NEW ZEALAND
Phone: +64 9 373 7599 ext 86830		Fax: +64 9 373 7000
Email:	d.scott at auckland.ac.nz

Graduate Officer, Department of Statistics
Director of Consulting, Department of Statistics