[R] Error propagation

[Ben Bolker]

Andrej-Nikolai Spiess <a.spiess <at> uke.uni-hamburg.de> writes:

> 
> Dear R-helpers,
> 
> I´m in the context of writing a general function for error propagation
> in R.
> There are somehow a few questions I would like to ask (discuss), as my
> statistical knowledge is somewhat restricted.
> Below is the function I wrote, the questions are marked.
> Many thanks in advance.

  Permutations: why do you want permutations and not
combinations?  (i.e., order is unimportant for covariances)

  Bartlett's test: seems like overkill.  I would say (in the
R spirit) that people should use some common sense if they
are trying to use the delta method/Gaussian error propagation

   It certainly makes sense to use the covariance, if you
have covariance information.

  I haven't looked at the rest of this terribly carefully --
you could look at ?deltamethod, ?deltavar in the emdbook
package (available from CRAN) as a comparison ...

  cheers
    Ben Bolker