[R] GLMM

[Prof Brian Ripley]

On Mon, 1 Nov 2004 clemens.tilke at web.de wrote:

> I have a problem concerning estimation of GLMM. I used methods from 3 different 
> packages (see program). 

You haven't really attributed the functions you use to particular
packages.  If this is glmm() from Jim Lindsey's packages then it was our
experience around the time that MASS was written that is often discordant
with other implementations/approaches.  As a result we have not used it
more recently.

> I would expect similar results for glmm and glmmML. The result differ in
> the estimated standard errors, however. I compared the results to MASS,
> 4th ed., p. 297. The results from glmmML resemble the given result for
> 'Numerical integration', but glmm output differs. For the intercept e.g.
> I have a standard error of 0.4354 from glmm and 0.5338 from glmmML. Any
> idea about this problem??

My main reaction is that the intercept is not an interesting parameter in
that problem and that small differences in standard error are not of any
practical consequence.  So what exactly is `this problem'?

> 
> Thanks in advance
> Clemens Tilke
> 
> 
> 
> library (MASS)
> data (bacteria)
> 
> contrasts (bacteria$trt) <- structure (contr.sdif (3), 
> dimnames = list (NULL, c ("drug", "enc")))
>  
> bacteria.pql <- glmmPQL (y ~ trt + I(week > 2), random = ~ 1 | 
> ID, family = binomial, data = bacteria)
> 
> summary (bacteria.pql)
> 
> library (repeated)
> 
> y1 <- 1 * (bacteria$y == "y")
> 
> bacteria1 <- data.frame (bacteria, y1 = y1)
> 
> bacteria.glmm <- glmm (y1 ~ trt + I(week > 2), nest = ID, data 
> = bacteria1, family = binomial)
> 
> summary (bacteria.glmm)
> 
> library (glmmML)
> 
> bacteria.glmmML <- glmmML (y1 ~ trt + I(week > 2), cluster = 
> bacteria1$ID, data = bacteria1, family = binomial)
> 
> summary (bacteria.glmmML)
> 
> bacteria.glmmML

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595