extRemes (version 2.1-3)

make.qcov: Covariate Matrix for Non-Stationary EVD Projections

Description

Create a matrix for use with pextRemes.

Usage

make.qcov(x, vals, nr = 1, ...)
is.qcov(x)

Value

An nr by np + 1 matrix is returned, where np is the number of parameters in the model. The last column is always “threshold” even if the model does not take a threshold (e.g., the GEV df), in which case the last column may be all NA, 0, or some other value depending on the vals argument.

Arguments

x

make.qcov: A list object of class “fevd” as output from fevd.

is.qcov: Any R object.

vals

Either a named list whose names match the fitted model parameter names, or may be “threshold”, a matrix or a numeric vector of length equal to the size of the resulting matrix.

nr

The number of rows desired in the resulting matrix. Only if vals is a vector. If vals argument is not a vector, the code will either fail or the argument will be ignored.

...

optional arguments to matrix (e.g., byrow=TRUE, depending on the order for vals, if a vector). Only used if vals is a vector.

Author

Eric Gilleland

Details

Simply sets up a matrix of parameter coefficients to be used by pextRemes. In particular, all parameters/thresholds that are constant (i.e., do not depend on covariate values) should have columns of all ones. Paramters/threshold that vary in a non-stationary model may have whatever values are of interest.

is.qcov performs some very simple tests to determine if an object is a proper qcov matrix or not. It is possible to have a matrix that is not a proper qcov matrix, but the returned value is TRUE. It is also possible to have a valid qcov object that id not appropriate for a particular model. Mostly this is an internal function.

See Also

pextRemes, fevd, erlevd

Examples

data(PORTw)
fit <- fevd(TMX1, PORTw, location.fun=~AOindex, units="deg C")
fit

v <- cbind(rep(1,4), c(1, -1, 1, -1), rep(1,4), rep(1,4))
v <- make.qcov(fit, vals=v, nr=4)
v

# cf.
v <- make.qcov(fit, vals=list(mu1=c(1, -1, 1, -1)))
v

# Or
v <- make.qcov(fit, vals=c(rep(1,4), c(1, -1, 1, -1), rep(1,8), rep(0,4)), nr=4)
v