glmmADMB package, built on the open-source
AD Model Builder platform, is
an R package for fitting
generalized linear mixed models (GLMMs).
Its capabilities include:
- a wide range of families (response distributions), including non-exponential families such as negative binomial (type 1 and 2), Beta, logistic, and truncated Poisson and negative binomial distributions as well as the standard exponential families (binomial, Poisson, Gamma, Gaussian).
- a wide range of link functions: log, logit, probit, complementary log-log, identity, inverse.
- Zero-inflation (currently only as a single constant term across all groups)
- Single or multiple random effects, including both nested and crossed effects
- Markov chain Monte Carlo (MCMC) summaries of uncertainty
In order to use glmmADMB effectively you should already be reasonably familiar with GLMMs, which in turn requires familiarity with (i) generalized linear models (e.g. the special cases of logistic, binomial, and Poisson regression) and (ii) 'modern' mixed models (those working via maximization of the marginal likelihood rather than by manipulating sums of squares).
Please visit the following webpages for more information about the glmmADMB package
(please note the latter is somewhat out of date,
although it may still contain useful information):
- First try
- If this fails (because you don't have the very latest version of R, or because R-forge is having a bad day), try
Arni Magnusson has uploaded glmmADMB 0.7.7 (16 July 2013) for R 3.0.x to a local repository:
If all else fails, contact the package maintainers.
Note that recent versions of
glmmADMB (>=0.7) require the
R2admb package as well; under normal circumstances this should be installed automatically when you install
glmmADMB, but if you run into trouble you should try solutions similar to those listed above.
- At present the binaries included in the
glmmADMB will not run on MacOS 10.5 (Leopard) or earlier.
If you encounter this problem, your choices are:
- Upgrade your system to a more recent version of MacOS (if possible);
glmmadmb from its TPL file on your machine. This will be a bit tricky if you are not reasonably experienced
- Download the full AD Model Builder source code
from the AD Model Builder download page and follow the directions for building AD Model Builder from source; you may need to install Xcode, and you may need to ask for help at
firstname.lastname@example.org. (Googling "admb macos 10.5" will be helpful as well, although it's possible that you will need the most recent version of the ADMB source code to compile
glmmadmb.tpl properly ...)
- find the
glmmadmb.tpl file in the
glmmADMB package directories and use ADMB to compile it to a binary
- copy the resulting binary to the
bin/macos64 directory as appropriate.
- Contact the maintainers to appeal for help and find out if
there any new developments in support for MacOS versions less than 10.6.
- A similar process may work for other unsupported
operating systems such as
Solaris, but in that case it's also probably a good idea to
contact the maintainers.
- We recommend the R mixed models list at
glmmADMB questions, although if you feel that your question is more AD Model Builder-oriented than R-oriented you may also want to try the AD Model Builder user's list.
- Current (fairly minimal) documentation/example for ADMB:
this is also accessible from within R (once
glmmADMB is installed) via
- The GLMM FAQ page gives general advice about GLMMs, although its content is slightly more oriented toward the
Note on new versions
New versions of glmmADMB (>0.6.4) have the following major changes:
The new release is somewhat slower (for the time being) than older (pre-0.5.2) versions: if you
have a desperate need for a copy of an old version, you can
download a source version and follow alternative #3 from the installation instructions above.
- new formula format, similar to that of the lme4 package, where random and fixed effects are specified as part of a single formula (random can also be specified separately, as in lme)
- multiple grouping variables (random effects) are allowed
- wider range of distributions and link functions supported (e.g. binomial with N> 1)