[R-sig-ME] Low intercept estimate in a binomial glmm

[Ben Bolker]

lborger <lborger at ...> writes:


> Cool! Might be nice to include into glmm.wikidot.com/faq?

  Be my guest -- it's a wiki after all ... (I've written most
of it, but there have been a few very welcome edits, in addition
to a whole bunch of commercial spam that I have to keep going
in and weeding ...)

> -----Original Message-----
> From: Jarrod Hadfield <j.hadfield at ...>

[snip]
> 
> I think the Diggle approximation is more accurate:
> 
> sd<-seq(0,4,length=100)
> x<-(-2.3776295)
> 
> Emu<-sapply(sd, function(sd){mean(plogis(rnorm(10000, x,sd)))})
> plot(Emu~sd) # simulated expectations
> c2<-((16*sqrt(3))/(15*pi))^2
> lines(plogis(x/sqrt(1+c2*sd^2))~sd, col="red")
> lines(plogis(x+0.5*sd)~sd, col="blue")
> 
> # approximations for the expectation