douconca (version 1.2.3)

predict.dcca: Prediction for double-constrained correspondence analysis (dc-CA)

Description

Prediction of traits from environment, environment from traits and response from trait and environment data.

Usage

# S3 method for dcca
predict(
  object,
  ...,
  type = c("envFromTraits", "traitsFromEnv", "response", "SNC", "CWM", "lc", "lc_traits"),
  rank = "full",
  newdata = NULL,
  weights = NULL,
  scaling = "symmetric"
)

Value

a matrix with the predictions. The exact content of the matrix depends on the type of predictions that are being made.

Arguments

object

return value of dc_CA.

...

Other arguments passed to the function (currently ignored).

type

type of prediction, c("envFromTraits", "traitsFromEnv", "response", "lc", "lc_traits") for environmental values, values of traits, response (expected abundance) and constrained scores for sites and species. "SNC" is equivalent with "envFromTraits". "CWM" is equivalent with "traitsFromEnv".

rank

rank (number of axes to use). Default "full" for all axes (no rank-reduction).

newdata

Data in which to look for variables with which to predict. For type = "envFromTraits" or "traitsFromEnv" or type = "lc_traits" or "lc", newdata is a data frame of trait and environmental values, respectively, which are used for the prediction or the calculation of scores. If omitted, fitted values are generated (use fitted.dcca instead). For type = "response", newdata is a list of two data frames with trait and environmental values in this order, e.g. list(traits = dataTraits, env = dataEnv).

weights

list of weights of species and of sites in newdata when type = "response", else ignored (default NULL yielding equal species and site weights, both summing to 1). Example: weights = list(species = c(100, 1, 1), sites = c(1, 1, 1, 1)), in that order, with traits of three new species in newdata[[1]] and environmental values (and levels of factors) of four new sites in newdata[[2]]. Species weights are scaled to a sum of one.

scaling

numeric (1,2 or 3) or character "sites", "species" or "symmetric". Default: "symmetric". Either site- (1) or species- (2) related scores are scaled by eigenvalues, and the other set of scores have unit weighted mean square or with 3 both are scaled symmetrically to weighted mean squares equal to the square root of eigenvalues. Negative values are treated as the corresponding positive ones by abs(scaling).

Details

Variables that are in the model but not in newdata are set to their weighted means in the training data. Predictions are thus at the (weighted) mean of the quantitative variables not included. Predictions with not-included factors are at the reference level (the first level of the factor).

For type = "response", many of the predicted values may be negative, indicating expected absences (0) or small expected response values.

With type = "traitsFromEnv" and newdata = NULL, predict gives the fitted mean traits, i.e. the fitted community weighted means. With type = "envFromTraits" and newdata = NULL, predict gives the fitted mean environment, i.e. the fitted species niche centroids (see fitted.dcca). See fitted.dcca.

Examples

data("dune_trait_env")

# rownames are carried forward in results
rownames(dune_trait_env$comm) <- dune_trait_env$comm$Sites

mod <- dc_CA(formulaEnv = ~ A1 + Moist + Mag + Use + Condition(Manure),
             formulaTraits = ~ SLA + Height + LDMC + Condition(Seedmass) + Lifespan,
             response = dune_trait_env$comm[, -1],  # must delete "Sites"
             dataEnv = dune_trait_env$envir,
             dataTraits = dune_trait_env$traits,
             verbose = FALSE)

# Ten 'new' sites with a subset of the variables in mod 
# X_lot will be ignored as it is not part of the model
newEnv <- dune_trait_env$envir[1:10, c("A1", "Mag", "Manure", "X_lot")]
newEnv[2,"A1"] <- 3.0
rownames(newEnv) <- paste0("NewSite", 1:10)

pred.traits <- predict(mod, type = "traitsFromEnv", newdata = newEnv)
head(pred.traits)


# Eight 'new' species with a subset of traits that are included in the model 
# Variable "L" will be ignored as it is not in the model 
newTraits <- dune_trait_env$traits[1:8, c("Species", "SLA", "LDMC", "L")]
newTraits[3,"SLA"]<- 18 
rownames(newTraits) <- paste("Species",LETTERS[1:8] )# or another meaningful name.

pred.env <- predict(mod, type = "envFromTraits", newdata = newTraits)
head(pred.env)


pred.resp <- predict(mod, type = "response", newdata = list(newTraits, newEnv),
                     weights = list(species = rep(1:2, 4), sites = rep(1, 10)))
colSums(pred.resp) # about alternating 0.8 and 1.6 (reflecting the new species weights)
rowSums(pred.resp) # about equal rowsums