R help - Nov 2001 - Help to conduct a random factor analysis with binomial response

Dear users of the R mailing list,

I am a ph.d. student in biology working on red deer in Norway, who would 
like to conduct an analysis with random factor where the response is 
binomially distributed. This cannot be conducted in S-plus, and I was told 
by others that it may be possible in R. However, I soon got into trouble 
which I hope you can help me to solve.

My model is on this form:
cbind(success,1-success) ~ variable 1 (numeric) + variable 2 (numeric) + 
variable 3 (factor) + variable 4 (factor). I have only one random factor, 
variable 5 (factor).

We assume that the analysis to be used is a non-linear mixed effects model, 
and hence use the nlme function in R. However, we failed to specify this 
model in a correct way. My question is therefore:

What is the syntax for this specific model (we got especially confused by 
the "start" argument)? If anyone could write the entire formula for
this
specific model we would be very greatful!


Thank you!


With regrads, Leif Egil Loe


.................................................................................................................................................
Leif Egil Loe                                        Privatadresse:
Avdeling for Zoologi                               c/o Hafting
Biologisk institutt                                 Hauges vei 31
Universitetet i Oslo                               N-0871 Oslo
Postboks 1050 Blindern
N-0316 Oslo

Tlf.    +47 22 85 72 94
Fax.  +47 22 85 46 05
Mob. +47 91 77 33 40
E-post. l.e.loe at bio.uio.no

.................................................................................................................................................

-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request at
stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._

Leif Egil Loe <l.e.loe at bio.uio.no> writes:
> I am a ph.d. student in biology working on red deer in Norway, who
> would like to conduct an analysis with random factor where the
> response is binomially distributed. This cannot be conducted in
> S-plus, and I was told by others that it may be possible in
> R. However, I soon got into trouble which I hope you can help me to
> solve.
This type of model is a generalized linear mixed model.

I think you would have to use the GLMMGibbs package in R to analyze
it.  You can't fit this type of model with the nlme package in R or in
S-PLUS.
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request at
stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._

Leif Egil Loe <l.e.loe at bio.uio.no> writes:
> Dear users of the R mailing list,
> 
> I am a ph.d. student in biology working on red deer in Norway, who
> would like to conduct an analysis with random factor where the
> response is binomially distributed. This cannot be conducted in
> S-plus, and I was told by others that it may be possible in
> R. However, I soon got into trouble which I hope you can help me to
> solve.
> 
> 
> My model is on this form:
> cbind(success,1-success) ~ variable 1 (numeric) + variable 2 (numeric)
> + variable 3 (factor) + variable 4 (factor). I have only one random
> factor, variable 5 (factor).
> 
[...]

I wrote a function in R that uses the EM algorithm for nonparametric
ML estimation in GLMs decribed by Aitkin (Biometrics 1999) first used
by Hinde ad Wood (1987). This is a generalization of the random
effects models for overdispersion in GLMs. The nice feature of this
approach is that you concurrently model normal as well as unknown
mixing distributions.  The algorithm alows for only one random effect.

The R function is based on the GLIM macro defined by AItkin and
Francis (1995 The GLIM Newsletter).

I would gladly run your data through for you.

Ross Darnell
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request at
stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._