mvabund (version 4.2.1)

anova.traitglm: Testing for a environment-by-trait (fourth corner) interaction by analysis of deviance

Description

Returns an analysis of deviance from a fourth corner model, as computed using traitglm, typically to test for an environment-by-trait interaction. Slowly! This function works via anova.manyglm, which uses row-resampling for inference, and it only applies to traitglm objects that have been fitted using the (default) manyglm function.

Usage

# S3 method for traitglm
anova(object, …, nBoot=99, resamp="pit.trap", test="LR",
                             block = NULL, show.time="all", bootID=NULL)

Arguments

object

A fitted object of class traitglm and class manyglm.

…

Additional traitglm objects, fitted using the formula argument.

nBoot

The number of bootstrap iterations. Default is 99 (NOTE: you should increase this for later runs!)

resamp,test,bootID

Arguments as in anova.manyglm, to control resampling method (resamp), test statistic (test) and whether or not a matrix of bootstrap resamples is manually entered (bootID).

block

A factor specifying the sampling level to be resampled. Default is resampling sites (which still involves passing a blocking variable to anova.manyglm, to keep all rows of the original abundance data together in resamples).

show.time

Whether to display timing information for the resampling procedure: this is advisable, as resampling fourth corner models many times can take a while. The default value "all" shows all timing information, "total" shows only the overall time taken for the tests, and "none" shows none.

Value

A list of values as returned by anova.manyglm, of which the most relevant element is table (the analysis of deviance table).

Warning

The comparison between two or more models by anova.traitglm will only be valid if they are fitted to the same dataset. This may be a problem if there are missing values and R's default of na.action = na.omit is used.

Details

There are two possible uses of this function, depending whether one traitglm object is specified or multiple objects.

If one traitglm object is specified, the anova.traitglm function returns a table summarising the statistical significance of the fourth corner terms in a model, that is, the interaction between environment and traits in predicting abundance across taxa and sites. All environment-by-trait interaction terms from the model are simultaneously tested.

If two or more nested traitglm objects are specified, and each has been fitted using a formula argument to the same set of datasets, then sequential test statistics (and P values) are returned for each additional model fit.

All traitglm models must be fitted using the manyglm function, which is its default behaviour, in order to access the anova.manyglm, which does most of the work. See anova.manyglm for details on how resampling is done, and options for arguments controlling the test statistic (via test) and the resampling method (via resamp). Because traitglm models are fitted by first vectorising the data into a univariate model, arguments such as p.uni and cor.type are redundant.

traitglm fits a single model to abundances across all sites and taxa at the same time, meaning the vector of abundances is typically pretty long, and the design matrix explaining how abundance varies across taxa and sites is typically pretty large. So resampling can take yonks. Hence the default number of resamples has been set at nBoot=99, but please consider increasing this once you have a sense for how long it will take to run (scales roughly linearly with nBoot).

References

Warton DI, Shipley B & Hastie T (2015) CATS regression - a model-based approach to studying trait-based community assembly, Methods in Ecology and Evolution 6, 389-398.

See Also

anova.manyglm,traitglm

Examples

# NOT RUN {
data(antTraits)

# we'll fit a small fourth corner model, to a subset of the antTraits data.
# first select only species present in at least 25% of plots:
abSum = apply(antTraits$abund>0,2,mean)
ab = antTraits$abund[,abSum>0.25]
tr = antTraits$traits[abSum>0.25,]

# now fit the fourth corner model, only as a function of a couple of traits and env variables:
ft=traitglm(ab,antTraits$env[,1:3],data.frame(tr$Weber,tr$Femur))
anova(ft,nBoot=39)
# Note you should refit with more bootstrap samples (e.g. 999), should take <2 minutes to run

# for an example using anova.traitglm for multiple fits, uncomment the following lines:
# ft2=traitglm(antTraits$abund,antTraits$env[,3:4],antTraits$traits[,c(1,3)],
#   formula=~1,composition=TRUE) #no fourth corner terms
# ft3=traitglm(antTraits$abund,antTraits$env[,3:4],antTraits$traits[,c(1,3)],
#   formula=~Shrub.cover:Femur.length+Shrub.cover:Pilosity,composition=TRUE) #shrub interactions
# ft4=traitglm(antTraits$abund,antTraits$env[,3:4],antTraits$traits[,c(1,3)],
#   formula=~Shrub.cover:Femur.length+Shrub.cover:Pilosity+Volume.lying.CWD:Femur.length+
#   Volume.lying.CWD:Pilosity, composition=TRUE) #all interactions only
# anova(ft2,ft3,ft4) # Shrub interactions not significant but CWD interactions are.
# }