glmulti: an R package for easy automated model selection with (Generalized) Linear Models
I have prepared the glmulti package to help investigators select the "best" models, or select variables of importance, in the framework of (Generalized) Linear Models (GLMs). The package provides an alternative to stepwise selection procedures, by comparing all possible candidate models and ranking them according to some Information Criterion (AIC, AICc or BIC). This approach guarantees that the best model is identified, and in addition allows to assess model selection uncertainty, and to perform multimodel inference/model averaging. Using glmulti is entirely automatic and is not more complicated than using standard functions like glm or lm.
The package features a Genetic Algorithm to deal with very large numbers of candidate models/variables. Contrary to most variable selection methods existing so far, glmulti can handle interactions between variables and can ensure that functional marginality is verified.
The package can be downloaded and installed from the Comprehensive R Archive Network (CRAN):
http://cran.mirrors.pair.com/web/packages/glmulti/index.html
An article explaining in details how the package works and containing many examples is currently being considered for publication.
The package is released under the GPL licence. The source JAVA code used in the package (3 classes in one package) can be downloaded here. Please email me for any question/suggestion!