VGAM: vcovvlm – R documentation

Become an expert in R — Interactive courses, Cheat Sheets, certificates and more!

vcovvlm

Calculate Variance-Covariance Matrix for a Fitted VLM or RR-VGLM or QRR-VGLM Object

Description

Returns the variance-covariance matrix of the parameters of a fitted vlm-class object or a fitted rrvglm-class object.

Usage

vcov(object, ...)
vcovvlm(object, dispersion = NULL, untransform = FALSE, complete = TRUE)
vcovqrrvglm(object, ...)

Arguments

`object`	A fitted model object, having class `vlm-class` or `rrvglm-class` or `qrrvglm-class` or a superclass of such. The former includes a `vglm` object.
`dispersion`	Numerical. A value may be specified, else it is estimated for quasi-GLMs (e.g., method of moments). For almost all other types of VGLMs it is usually unity. The value is multiplied by the raw variance-covariance matrix.
`untransform`	logical. For intercept-only models with trivial constraints; if set `TRUE` then the parameter link function is inverted to give the answer for the untransformed/raw parameter.
`complete`	An argument that is currently ignored. Added only so that `linearHypothesis` can be called.
`...`	Same as `vcov`.

Details

This methods function is based on the QR decomposition of the (large) VLM model matrix and working weight matrices. Currently vcovvlm operates on the fundamental vlm-class objects because pretty well all modelling functions in VGAM inherit from this. Currently vcovrrvglm is not entirely reliable because the elements of the A–C part of the matrix sometimes cannot be computed very accurately, so that the entire matrix is not positive-definite.

For "qrrvglm" objects, vcovqrrvglm is currently working with Rank = 1 objects or when I.tolerances = TRUE. Then the answer is conditional given C. The code is based on model.matrixqrrvglm so that the dimnames are the same.

Value

Same as vcov.

Note

For some models inflated standard errors can occur, such as parameter estimates near the boundary of the parameter space. Detection for this is available for some models using hdeff.vglm, which tests for an Hauck-Donner effect (HDE) for each regression coefficient. If the HDE is present, using lrt.stat.vlm should return more accurate p-values.

Author(s)

Thomas W. Yee

Examples

ndata <- data.frame(x2 = runif(nn <- 300))
ndata <- transform(ndata, y1 = rnbinom(nn, mu = exp(3+x2), size = exp(1)),
                          y2 = rnbinom(nn, mu = exp(2-x2), size = exp(0)))
fit1 <- vglm(cbind(y1, y2) ~ x2, negbinomial, data = ndata, trace = TRUE)
fit2 <- rrvglm(y1 ~ x2, negbinomial(zero = NULL), data = ndata)
coef(fit1, matrix = TRUE)
vcov(fit1)
vcov(fit2)

VGAM

Vector Generalized Linear and Additive Models

v1.1-5

GPL-3

Authors

Thomas Yee [aut, cre], Cleve Moler [ctb] (author of several LINPACK routines)

Initial release

2021-01-13