fn.one.rep.y: Perform a single Metropolis-Hastings step
In jrlewi/brlm: Bayesian Restricted Likelihood Methods

Description Usage Arguments Details Value

Performs a single Metropolis-Hastings step for the restricted likelihood method

fn.one.rep.y(
  y.curr,
  Beta,
  sig,
  l1obs,
  s1obs,
  X,
  log.prop.den.curr,
  proj,
  fn.psi,
  fn.chi,
  psi,
  scaleEst,
  maxit = 400
)

fn.one.rep.y2(
  y.curr,
  Beta,
  sig,
  l1obs,
  s1obs,
  X,
  log.prop.den.curr,
  proj,
  Qt,
  fn.psi,
  fn.chi,
  psi,
  scaleEst,
  maxit = 400
)

`y.curr`	current data vector
`Beta`	vector of regression coefficient parameters
`sig`	sigma-the standard deviation parameter
`l1obs`	conditioning statistic-estimate of Beta from observed data
`s1obs`	conditioning statistic-estimate of sigma from observed data
`X`	regression design matrix (i)
`log.prop.den.curr`	value of the proposal density for the current data vector (free of other parameters so it can be saved from iteration to iteration to save computation time)
`proj`	the projection matrix onto the deviation space (aka the least squares residual space)
`fn.psi`	custom psi function returning psi(x) or psi'(x) defining M-estimator for location ; default is fn.psi.huber
`fn.chi`	custom chi function returning chi(x) or chi'(x) defining M-estimator for scale; default (and currently only capability) is fn.chi.prop2
`psi`	psi function returning psi(x)/x. Must match fn.spi, but currently no warning is thrown.
`scaleEst`	'Huber' (default and currently only capability).
`maxit`	max iterations for the estimation of the summary statistics
`Qt`	Q transpose where Q is the orthonormalized X

Two equivalent versions, the second uses fn.attenuation2 and is faster. This function is designed for use within an MCMC sampler. Many of the inputs are iteration dependent. The input log.prop.den.curr is designed as a time saver. Since this is free of other parameters it can be saved from iteration to iteration to save computation time. Note the use of both fn.psi (custom in the brlm package) and psi (from the MASS package). This is poor coding and should be corrected.

list y.curr: the new (if accepted) or same (if rejected) data vector, a: indicator of acceptance of proposed value 1 if accepted, 0 if rejected, log.prop.den.curr: proposal density on log scale of the returned data vector, rat: MH ratio, log.rat1: log ratio of proposal densities

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