[R] glm specification where response is a 2col matrix

Uwe Ligges ligges at statistik.tu-dortmund.de
Tue Jan 11 18:42:32 CET 2011


when I apply a glm() model in two ways,
first with the response in a two column matrix specification with 
successes and failures

y <- matrix(c(
     5, 1,
     3, 3,
     2, 2,
     0, 4), ncol=2, byrow=TRUE)

X <- data.frame(x1 = factor(c(1,1,0,0)),
                 x2 = factor(c(0,1,0,1)))

glm(y ~ x1 + x2, data = X, family="binomial")

second with a model matrix that full rows (i.e. has as many rows as real 
observations) and represents identical data:

Xf <- data.frame(x1 = factor(rep(c(1,1,0,0), rowSums(y))),
                  x2 = factor(rep(c(0,1,0,1), rowSums(y))))
yf <- factor(rep(rep(0:1, 4), t(y)))

glm(yf ~ x1 + x2, data = Xf, family="binomial")

we will find that the number of degrees of freedom and the AIC etc. 
differ --  I'd expect them to be identical (as the coefficient estimates 
and such things are).

maybe I am confused tonight, hence I do not file it as a bug report 
right away and wait to be enlightened ...

Thanks and best wishes,

More information about the R-help mailing list