cocorresp (version 0.4-1)

permutest.coca: Permutation test for predictive co-correspondence analysis models

Description

A permutation test for predictive co-correspondence analysis models to assess the significance of each CoCA ordination axes.

Usage

# S3 method for coca
permutest(x, R0 = NULL, permutations = 99,
               n.axes = x$n.axes, verbose = TRUE, …)
# S3 method for permutest.coca
summary(object, …)

Arguments

x

an object of class "predcoca".

R0

row weights to use in the analysis. If missing, the default, these are determined from x.

permutations

the number of permutations to perform.

n.axes

The number of axes to test. Defaults to the number of axes stated in x$n.axes.

verbose

if TRUE, the default, print information on the progress of the permutation test procedure.

object

an object of class "permutest.coca".

…

arguments to be passed to other methods.

Value

A list with the following components:

pval

a vector of P-values for each ordination axis.

permstat

a vector of values for the test statistic for each axis.

total.inertia

the total inertia in the response matrix.

inertia

a vector containing the residualised inertia. This is the total inertia in the response after removing the inertia explained by all previous axes. For the first CoCA axis this is, by definition, the total inertia in the response.

fitax

a vector containing the amount of inertia in the response matrix explained by each ordination axis.

pcent.fit

a vector containing the fit of each axis to the response as a percentage of the total inertia (variance).

n.axes

the number of axes in the ordination.

call

the matched call.

Warning

This function is slow. Beware setting argument permutations higher than the default. Determine how long it takes for the default 99 permutations to complete before going crazy and asking for thousands of permutations - you've been warned, have a good book to hand.

Details

An alternative approach to cross-validation (see crossval) to select the number of axes to retain in a predictive co-correspondence analysis is to test the statistical significance of each ordination axis using permutation tests.

The test statistic used is the F-ratio based on the fit of the first axis to the response data (ter Braak and Smilauer 2002). The second and subsequent axes are tested by treating previous axes as co-variables.

To be precise, this approach does not test the significance of SIMPLS axes, but those of NIPALS-PLS axes (ter Braak and de Jong 1998).

References

ter Braak, C.J.F. and de Jong, S. (1998) The objective function of partial least squares regression. Journal of Chemometrics 12, 41--54.

ter Braak, C.J.F and Schaffers, A.P. (2004) Co-Correspondence Analysis: a new ordination method to relate two community compositions. Ecology 85(3), 834--846.

ter Braak, C.J.F. and Smilauer, P. (2002) Canoco reference manual and CanoDraw for Windows user's guide: software for canonical community ordination. Version 4.5. New York: Microcomputer Power.

See Also

coca, for the model fitting function, crossval, for a leave-one-out cross-validation procedure, which is the preferred way to select axes in a predictive co-correspondence analysis.

Examples

# NOT RUN {
## load some data
data(beetles)
data(plants)

## log transform the bettle data
beetles <- log(beetles + 1)
## predictive CoCA using SIMPLS and formula interface
bp.pred <- coca(beetles ~ ., data = plants)

## should retain only the useful PLS components for a parsimonious model
# }
# NOT RUN {
## Leave-one-out crossvalidation - this takes a while
crossval(beetles, plants)
## so 2 axes are sufficient
# }
# NOT RUN {
## permutation test
## (Testing the first 2 axes & only 50 perms for speed.)
bp.perm <- permutest(bp.pred, permutations = 50, n.axes = 2)
bp.perm

# }