restrictedBayesLm: MCMC algorithm for Restricted Likelihood Bayesian Linear...
In jrlewi/brlm: Bayesian Restricted Likelihood Methods

Description Usage Arguments Details Value

Fits the linear regression model β~N(μ_0, Σ_0), σ^2~IG(a0, b0), y~N(Xβ, σ^2 I)

restrictedBayesLm(
  y,
  X,
  regEst = "Huber",
  scaleEst = "Huber",
  mu0,
  Sigma0,
  a0,
  b0,
  sigma2Int,
  nkeep = 10000,
  nburn = 1000,
  maxit = 400
)

`y`	vector of repsonse
`X`	design matrix for regression. If intercept is desired, a column of 1's is needed
`regEst`	The estimator of β to condition on. Current options available are `Huber` and `Tukey` for Huber's and Tukey's (bisquare) estimators, respectively.
`scaleEst`	The estimator of σ to condition on. Currently Huber's proposal 2 is the only option available. Specified with `Huber`.
`mu0`	prior mean for beta
`Sigma0`	prior var-cov matrix for β
`a0`	prior shape parameter for σ^2
`b0`	prior scale parameter for σ^2
`nkeep`	number of iterations to keep
`nburn`	number of iterations to toss
`maxit`	maximum number of iterations for use in `rlm` when computing conditioning statistic during each proposal within the MCMC step for [y \| rest, observed statistics]

Uses Gibbs sampling to sample from the restricted posterior under the above linear regression model conditioned on estimators regEst and scaleEst for β and σ, respectively. This is just the Gibbs sampler for the full posterior (implemented in brlm::bayesLm augmented with a step to sample new data given all of the parameters and the observed conditioning statistics

list with mcmc sample, mean fitted values, robust regression object from rlm, and observed conditioning statistics.

jrlewi/brlm documentation built on March 17, 2021, 1:10 a.m.