Full-subsets information-theoretic approaches are increasingly used to explore predictive power and variable importance when a wide range of candidate predictors are being considered. This package provides functions that can be used to construct, fit, and compare a complete model set of possible ecological or environmental predictors for a given response variable of interest. Models are based on Generalized Additive Models (GAMs) and build on the MuMIn package. Advantages include the capacity to fit more predictors than there are replicates, automatic removal of models with correlated predictors, and support for model sets that include interactions between factors and smooth predictors, as well as smooth-by-smooth interactions via te().
Details
Full subsets information theoretic approaches are becoming an increasingly popular tool for exploring predictive power and variable importance where a wide range of candidate predictors are being considered.
This package provides simple function(s) that can be used to construct, fit and compare a complete model set of possible ecological or environmental predictors, given a response variable of interest. The function(s) are based on Generalized Additive Models (GAM) and builds on the MuMIn package.
Advantages include the capacity to fit more predictors than there are replicates, automatic removal of models with correlated predictors, and model sets that include interactions between factors and smooth predictors, as all as smooth interactions with other smooths (via te).
The function(s) takes a range of arguments that allow control over the model set being constructed, including specifying cyclic and linear continuous predictors, specification of the smoothing algorithm used and the maximum complexity allowed for smooth terms.
The full subsets analysis can be carried out via one of two alternative methods allowed in the package.
The first is through a direct call to full_subsets_gam (this is the original function). This function both constructs and fits the complete model set, based on the user supplied input. This function requires that all model fits are saved, and is therefore not suitable for extremely large models sets, as these will cause issues with memory. This method may be superceded in future versions of FSSgam, so for any new project please use the second method.
The second method is via a call to generate_model_set followed by a second call to fit_model_set. This pair of functions splits the process of generating the model set from actually fitting and extracting the relevant model data. This method is useful for large model sets, because it allows the model set to be interrrogated before fitting and also optionally allows model fit data to not be saved, thus alleviating memory issues.
The use of the function(s) is demonstrated via case studies that highlight how appropriate model sets can be easily constructed, and the broader utility of the approach for exploratory ecology. Please see the case study files on github for usage examples at https://github.com/beckyfisher/FSSgam
References
Fisher R, Wilson SK, Sin TM, Lee AC, Langlois TJ (2018) A simple function for full-subsets multiple regression in ecology with R. Ecology and Evolution https://onlinelibrary.wiley.com/doi/abs/10.1002/ece3.4134
Author
Rebecca Fisher (Australian Institue of Marine Science)
Maintainer: Rebecca Fisher r.fisher@aims.gov.au