Skip to content
Prev 11003 / 63424 Next

glm offset and interaction bugs (PR#4941)

On Tue, 4 Nov 2003 charlie@stat.umn.edu wrote:

            
Only one bug, i think.  There is certainly a bug in handling offset() in
formulas with interactions (which is presumably fairly recent, since the C
code for offsets dates only from January this year).

The latter two models, though, are the same AFAICS, or more precisely are
reparametrisations of the same underidentified model. The only way to make
+ commutative in this example would be to have some arbitrary rule based
on the name of the variable for which parameters to drop, and that would
have the even more unintuitive effect that changing the name of a variable
would affect the output.
Thomas Lumley			Assoc. Professor, Biostatistics
tlumley@u.washington.edu	University of Washington, Seattle