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

[Spencer Graves]

Hi, Peter: 

      Thanks.  The help page on GLMM in lme4 0.6-1 2004/05/31 mentions 
GLMM(formula, family, data, random, ...) with additional arguments 
subset, method, na.action, control, and "model, x logicals".  I may try 
reading the source code. 

      On the other hand, my need is sufficiently specialized that I may 
just program the log(likelihood) for that specific model.  Then I can 
make contour and perspective plots of the log(likelihood) surface to 
examine thereby the adequacy of Wald's approximation for different 
parameterizations, as well as feed it to "optim" for estimation.  [My 
application requires binomial(link="cloglog") not "logit";  I used 
"logit" in the simulation below, because it is more commonly known and 
understood.] 

      Best Wishes,
      Spencer Graves

Peter Dalgaard wrote:

>Spencer Graves <spencer.graves at pdf.com> writes:
>
>  
>
>>Data: DF
>>  log-likelihood:  -55.8861
>>Random effects:
>>  Groups Name        Variance   Std.Dev.
>>  smpl   (Intercept) 1.7500e-12 1.3229e-06
>>
>>Estimated scale (compare to 1)  3.280753
>>
>>Fixed effects:
>>             Estimate Std. Error z value Pr(>|z|)
>>(Intercept) 0.148271   0.063419  2.3379  0.01939
>>
>>Number of Observations: 10
>>Number of Groups: 10
>>    
>>
>
>The only information you have about sigma is in the overdispersion
>of y, so you probably cannot both have a scale parameter and a random
>effect of sample in there. Doug did say something about fixing the
>scale in GLMM, but I forget whether he had implemented it or planned
>to...
>
>  
>