[Rd] Apparant bug in binomial model in GLM (PR#13434)

Daniel Sabanés Bové daniel.sabanesbove at gmx.net
Wed Jan 7 13:41:54 CET 2009

> There appear to be a bug in the estimation of significance in the binomial model
> in GLM. This bug apparently appears when the correlation between two variables
> is to strong.
> Such as this dummy example
> c(0,0,0,0,0,1,1,1,1,1)->a
> a->b
> m1<-glm(a~b, binomial)
> summary(m1)
> It is sufficient that all 1's correspond to 1's such as this example
Isn't this just the well-known failure of logistic regression (non-existence of maximum likelihood estimator) for linearly separable data? (b > 0.5 => a == 1)
> c(0,0,0,0,0,1,1,1,1,1)->a
> c(0,0,0,0,1,1,1,1,1,1)->c
> m1<-glm(a~c, binomial)
> summary(m1)
This is not linearly separable, but almost, so the standard errors will be very large.

More information about the R-devel mailing list