varianceestimator.sandwich: Estimators for the PIM parameter variances

In pimold: Probabilistic Index Models

Description

Estimators for the PIM parameter variances

Usage

varianceestimator.sandwich(Uforposandwich = Uforposandwich.default)

varianceestimator.H0(Uforposandwich = Uforposandwich.fakeH0)

varianceestimator.glm(control = list())

varianceestimator.bootstrap(D = 500, keep.posetbs = FALSE)

Arguments

Uforposandwich	function like Uforposandwich.default (the default) to calculate the necessary parts for posandwich.estimator.
control	See glm.fit.
D	Number of bootstrap repetitions (defaults to 500).
keep.posetbs	If this is TRUE (not the default), the morevarfitinfo will be a list of two items: allcoefs containing the fitted coefficients, and posetbs, containing which pseudoobservations were bootstrapped. In both, the bootstrap repetitions are in the columns.

Details

These functions estimate the variances of the parameter estimates of the PIM. Each provides a solution under different circumstances and some may be more accurate / correct / performant depending on the specific model.

For the different implementations, morevarfitinfo contains:

1. varianceestimator.sandwichThe return value of the Uforposandwich call. This estimate is the generally valid sandwich estimator for the (co)variance.

2. varianceestimator.H0The return value of the Uforposandwich call. This estimate is the estimator for the (co)variance under H0.

3. varianceestimator.glmThe return value of the glm call. This estimate is only valid if the pseudo-observations are somehow independent and the linkfunction is identity.

4. varianceestimator.bootstrapA matrix holding the coefficient estimates over all bootstrap repetitions. The columns represent the repetitions

Value

These functions (varianceestimator.*) each return a function themselves. The returned function should have five parameters (estimationresult, the parameter estimates themselves (result of estimation.*); pfd, an object of class pimfitdata ; link, the name of the link function; estimator, the estimator used to generate estimationresult; verbosity, see elsewhere) and should itself return a list of two items:

vcov	The estimated covariance matrix
morevarfitinfo	Implementation specific information on the variance estimation

pimold documentation built on May 2, 2019, 5:50 p.m.