On Thu, 26 Jul 2001 Philippe.Guiblin@ons.gov.uk wrote:
> Does anyone know if there is any possibility to get Splus code able to
> perform a logistic regression model with/without random effects.
Without, glm(, family=binomial()) for the two-class class, plus lots of
interpretations (e.g. multiple logistic regression in multinom in library
nnet).
With random effects, it depends what you mean. In full generality it is
research problem, as the integration is difficult and MCMC methods are
often used. (Jonathan Myles had a GLMMGibbs package, but I think that is
currently only available for R.) If the random effect is only a
per-individual intercept, then numerical quadrature can be used. (Jim
Lindsey for example has a glmm() function in that extremely restricted
sense.) In between are PQL and other approximations: see Gordon Smyth's
code at
http://www.maths.uq.edu.au/~gks/s/index.html
(function reglm) for one example. Searching the S-news archive should
reveal more: this has come up before.
>
> Thanks,
>
> Philippe Guiblin
> Office for National Statistics
> 1 Drummond Gate
> SW1V 2QQ london
> UK
[Advert deleted.]
--
Brian D. Ripley, ripley@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 272860 (secr)
Oxford OX1 3TG, UK Fax: +44 1865 272595
|