funspace (version 0.2.2)

funspaceNull: Null models in functional space

Description

Comparing the amount of occupied functional space against null models

Usage

funspaceNull(
  funspace,
  nrep = 100,
  alter = "greater",
  null.distribution = "multnorm",
  verbose = TRUE
)

Value

funspaceNull The function returns the list containing all the simulated datasets, the area of the observed trait space, the mean value of the area for the null model (calculated across iterations), the p-value of the difference between observed and simulated trait space, as well as a standardized effect size of the difference between observed trait space and mean null model areas. This output is reported together with the output of funspace.

Arguments

funspace

An object of class funspace

nrep

numericThe number of generated null surfaces

alter

character. The hypothesis to be tested when comparing the observed trait space against the null model. Options are 'greater', 'less', and 'two-sided'. See specification of the as.randtest function in the ade4 R package

null.distribution

character. Data distribution for null model building. Available options are 'multnorm' and 'uniform' to generate data with a multivariate normal or uniform distribution, respectively.

verbose

logical. Do you want to information about the progress of the null model to be written to the console?

Details

funspaceNull The function tests for the statistical difference between the size (functional richness) of the considered TPD, obtained using the funspace function, against a vector of functional richness values generated using null models (see below) across a user-defined number of iterations. Two null models are currently available for testing. One generates data with a multivariate normal distribution, creating a dataset with normally distributed variables having the same mean and covariance than the observations used to build the functional space (see Carmona et al. 2021). This null model returns a theoretical TPD where some trait combinations (those around the mean of the trait space axes, thus towards the center of the null trait space) are more likely than others (i.e., this null model resembles an ellipse). The other null model generates a dataset with variables following a uniform distribution (see null model 1 in Diaz et al. 2016), creating a distribution where all trait combinations within the range of the original observations are equally possible (i.e., the approximate shape of this null model is a rectangle). Note that the function does not work for funspace objects that are based on a TPDs object created using the package TPD

References

CP Carmona, et al. (2021). Fine-root traits in the global spectrum of plant form and function. Nature 597, 683–687 S Diaz, et al. (2016). The global spectrum of plant form and function. Nature 529, 167–171

Examples

# 1. PCA space, multivariate model (see Carmona et al. 2021, Nature)
x <- princomp(GSPFF)
funtest <- funspace(x = x, PCs = c(1, 2), threshold = 0.95)
funtestNull <- funspaceNull(funtest, null.distribution = 'multnorm', nrep = 1000)
summary(funtestNull)

#'# 2. Two raw traits and uniform distribution (see Diaz et al. 2016, Nature)
x <- GSPFF[, c("ph", "sla")]
funtest <- funspace(x = x, threshold = 0.95)
funtestNull <- funspaceNull(funtest, null.distribution = 'uniform', nrep = 1000)
summary(funtestNull)