Message-ID: <7A3A13F416B40842BD2C1753E044B359B098C6@CASEVS02.cas.anu.edu.au>
Date: 2003-05-28T00:35:47Z
From: Simon Blomberg
Subject: Bradley Terry model and glmmPQL
aargh. "nb" should be "a" for each of the model formulae. This is not the source of the error. sorry for any confusion.
Simon.
>
> fit <- glm(cbind(success, failure) ~ c + b + nb -1,
> family=binomial, data=dat)
>
> which works fine. I can also include the age factor, which
> also seems to work ok:
>
> fit2 <- glm(cbind(success, failure) ~ c + b + nb +
> as.factor(age) -1, family=binomial, data=dat)
>
> Now, since each rater performs ratings on each of the 3
> treatment combinations, I was interested in including rater
> as a random factor. My naive method was to use glmmPQL from
> library MASS:
>
> fit3 <- glmmPQL(cbind(success, failure) ~ c + b + nb -1,
> random = ~1|rater, family=binomial, data=dat)
>
> However, I get the following error:
>
> iteration 1
> Error in MEEM(object, conLin, control$niterEM) :
> Singularity in backsolve at level 0, block 1