R: Standard Error Estimate

vcovfestRidge {eshrink}

R Documentation

Standard Error Estimate

Description

Computes an estimate of the variance-covariance matrix for an 'fLoss' ridge estimator

Usage

vcovfestRidge(
  fLoss,
  lambda,
  XtX,
  postBeta,
  postSigma2,
  penalize,
  ind = 1,
  version = c("varExp", "full")
)

Arguments

`fLoss`	A matrix of loss function values to be minimized. Assumed structure is the columns correspond to different values of penalty parameter and rows correspond to points in a posterior sample of (beta, sigma).
`lambda`	The sequence of penalty parameter values corresponding to the columns of `fLoss`.
`XtX`	Cross product of the design matrix.
`postBeta`	Matrix containing the posterior sample of beta values. Assumed to be n-by-p, where n is number of samples (and equal to number of rows in fLoss) and p is number of regression parameters in model.
`postSigma2`	Vector containing the posterior sample of variance parameters. Should have same length as postBeta.
`penalize`	Vector indicating which variables are penalized in the regression model. See details for `estRidge` for further details.
`ind`	Numerical or logical vector indicating which elements of the variance matrix should be returned. Defaults to the (1,1) element
`version`	Character string indicating which version of standard error to compute. 'varExp' or 'full', corresponding to the variance of the conditional mean of the estimator or that plus the expected value of the conditional variance. In practice, the latter is often too large.

Details

Computes a standard error estimate for an 'fLoss' estimator, where 'fLoss' is typically fMSE or fMBV. Approximates the variance of the estimator using the the variance conditional on the observed data (i.e. using the posterior distribution of parameters). Currently, two different versions are available.

Author(s)