Dear list members,
I’m having some problems understanding why drop1() and anova() gives
different results for *Gaussian* glm models. Here’s a simple example:
d = data.frame(x=1:6,
group=factor(c(rep(A,2), rep(B, 4
l = glm(x~group, data=d)
Running the following code
I am somewhat surprised that _anything_ sensible comes out of anova.glm(l,
test=Chisq). I think it is mostly expected that you use F tests for that case.
What does seem to come out is the same as for drop1(l, test=Rao), which gives
the scaled score test, which would seem to be equivalent to
2 matches
Mail list logo