[R] GLMM(..., family=binomial(link="cloglog"))?

[Douglas Bates]

Göran Broström <gb at stat.umu.se> writes:

> I think the "weights=n" is the problem, i.e., you summarize Bernoulli
> data to Bin(100, p) data, and that gives a completely different estimate of
> the variance of the random effect. (This looks as an error in lme4 to me,
> or am I missing something? Doug?) Really, the two ways of representing data
> should give equivalent analyses, but it doesn't. The same phenomenon
> appears in 'glm', i.e. without random effects, but only for the residual
> sum of squares, df, and AIC. 

It is indeed an error in lme4.  Thanks for picking that up, Goran.
Fixing it is now on the "To Do" list.