This should probably be submitted eventually as a wishlist to R-core,
as it requires (minor) changes to base R, but I thought I would float it
here first ...
In a package, I construct glm()-like model objects that have 'family'
components based on the GLM family objects (binomial(), poisson(),
Gamma(), etc.). I have written a 'simulate' method for these objects
that has basically had to re-code much of what's already contained in
the family()$simulate methods, because they are just a little bit too
hardcoded. I can't just add list elements to my object, because it's
represented by an S4 class .......
For example, poisson()$simulate contains references to
* object$prior.weights -- could be replaced by weights(object) ?
* fitted(object) -- I would like to be able to substitute my own "mu"
value here, but I can hack around it (in a perfect world I would add a
third optional argument:
$simulate <- function(object,nsim,ftd=fitted(object)) {
...
}
binomial()$simulate contains references to the above two components,
as well as a reference to object$model (which could be replaced by
model.frame(object)).
inverse.gaussian()$simulate has a reference to
summary(object)$dispersion (this wouldn't be too hard)
The required changes would be fairly minor, and as far as I can tell
completely backward compatible. Of course I don't know if there's much
demand for it ... it would certainly be helpful for me, I don't know if
there is latent demand out there ...
thoughts?
Ben Bolker
