[R] diagnostic information in glm. How about N of missing observations?

[Barry Rowlingson]

Frank E Harrell Jr wrote:

> Fitting functions in the Design package tell you how many observations
> were deleted due to each variable in the model.  They generalize the
> na.action component stored in the fit object.

  Surely accessing components of an object is generally a bad idea on 
O-O design, and there ought to be a _method_ that returns which 
observations have been omitted. Perhaps an 'omitted' function?

  Currently the $na.action component of lm (and glm) objects is an 
object of class "omit" (or possibly 'exclude', or even something else), 
which doesn't even have a print method of its own.

  Bizarrely it has an 'naprint' method:

 > naprint(fit$na.action)
[1] "2 observations deleted due to missing"

  But that returns a string which you'd have to cut up to get the number 
2 out of it. And why isn't that the 'print' method?

  naprint.lm remains to be written, so this reverts to na.default:

 > naprint(fit)
[1] ""

  Which at the very least is misleading...

  And relying on the $na.action object to store the omitted rows is 
asking for trouble - this element is already being used for different 
classes of object, so who knows what might be in it in the future...

  Is all this due for a rewrite for R 2.0.0?

Baz